Lithography Process Control Using Pre-Exposure Metrology and AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lithographic processes face challenges in accurately aligning and patterning substrates due to distortions in the wafer grid, which affect the placement of features and yield in semiconductor manufacturing, particularly at sub-micron scales, where traditional measurement methods are insufficient for high-order corrections.
Innovation Solution
A method involving machine learning to process pre-exposure metrology data, decomposing it into components that correlate with performance metrics like overlay, and applying a trained model to predict and correct distortions, enabling more precise alignment and patterning by filtering out correctable components and using dense external metrology data for improved substrate handling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional measurement methods are used to measure wafer grid distortions, then the measurement process is simple and fast, but the measurement precision is insufficient to correct higher-order distortions accurately
Solution Approach 1:
The patent introduces an intermediary machine learning model that mediates between the raw metrology data and the final overlay correction. The model processes pre-exposure metrology data (containing wafer grid distortion information) and outputs predicted overlay errors, serving as a computational intermediary that bridges measurement and correction without requiring direct complex measurement of higher-order distortions
Solution Approach 2:
The patent applies preliminary action by measuring and characterizing wafer grid distortions before the lithography exposure process. The pre-exposure metrology data captures distortion information in advance, allowing the machine learning model to predict and correct overlay errors before the actual patterning occurs, thereby improving measurement precision without adding complexity to the exposure process itself
2Manufacturing precision
If machine learning models are trained to predict overlay errors from pre-exposure metrology data, then the manufacturing precision improves, but the device complexity increases
Solution Approach 1:
The machine learning model implements self-service by automatically learning the complex relationships between pre-exposure metrology data and overlay errors from training data. Once trained, the model autonomously predicts overlay errors for new substrates without requiring manual intervention or complex real-time calculations, thereby improving manufacturing precision while keeping the operational complexity manageable
Solution Approach 2:
The patent replaces complex mechanical measurement and correction systems with a computational machine learning model. Instead of using more complex hardware to directly measure and correct higher-order distortions, the system uses software-based AI algorithms that process existing metrology data to predict and correct overlay errors, achieving improved precision with controlled complexity
3Manufacturing precision
If pre-exposure metrology data is processed and decomposed into components, then the manufacturing precision improves, but the loss of time increases
Solution Approach 1:
The patent applies preliminary action by performing the time-consuming data decomposition and model training in advance, before actual production. The machine learning model is pre-trained on historical data containing decomposed components, allowing rapid prediction during manufacturing without requiring real-time decomposition, thus improving patterning accuracy while minimizing processing time during production
Solution Approach 2:
The patent segments the pre-exposure metrology data into distinct components (such as wafer grid distortion components, alignment mark position components, etc.) that can be independently processed and correlated with overlay errors. This segmentation allows the machine learning model to efficiently process and interpret data components separately, improving manufacturing precision while reducing the computational burden and time required during actual production
Data Source
AI summary
A method for determining a correction relating to a performance metric of a semiconductor manufacturing process, the method including: obtaining a set of pre-process metrology data; processing the set of pre-process metrology data by decomposing the pre-process metrology data into one or more components which: a) correlate to the performance metric; or b) are at least partially correctable by a control process which is part of the semiconductor manufacturing process; and applying a trained model to the processed set of pre-process metrology data to determine the correction for the semiconductor manufacturing process.