Machine Learning Controller for Lithography System Drift Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing lithography systems are ineffective in correcting system drift due to a non-optimal machine learning controller that fails to accurately characterize external and internal elements affecting accuracy, leading to unreliable overlay predictions and corrections.
Innovation Solution
A lithography system equipped with a machine learning controller trained on post-exposure measurement data to adjust the lithography apparatus, track unit, and control unit, allowing for accurate monitoring and correction of system drift based on measured properties of multiple substrates, thereby optimizing overlay and critical dimension control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a linear formula with tool specific coefficients is used to predict overlay from alignment parameters, then the control method is simple and easy to implement, but the prediction reliability is insufficient because the formula cannot properly characterize all external and internal elements that affect accuracy
Solution Approach 1:
The patent transforms the control approach by changing from a static linear formula to a dynamic machine learning model that continuously adapts parameters based on measured pattern properties. The machine learning controller learns optimal correction parameters from actual measurement data, enabling the system to capture complex non-linear relationships between alignment parameters and overlay errors that a fixed linear formula cannot represent.
Solution Approach 2:
The patent implements a feedback mechanism where the machine learning controller uses measured pattern properties from the metrology unit to continuously refine and update its predictions. The system feeds back actual measurement results to the machine learning model, allowing it to adapt to system drift and improve overlay prediction reliability over time, rather than relying on a predetermined linear formula.
2Productivity
If alignment parameters are used to calculate and correct overlay in advance, then correction can be applied on a substrate-to-substrate basis, but system drift cannot be effectively corrected because real-time parameters are mixed with systematic drift parameters
Solution Approach 1:
The patent segments the overlay error into two distinct components: substrate-specific random variations captured by alignment parameters and systematic drift captured by measured pattern properties. The machine learning controller separately processes these components, applying alignment parameter corrections for individual substrates while using pattern measurement feedback to correct systematic drift affecting multiple substrates, thereby improving both productivity and precision.
Solution Approach 2:
The patent introduces the machine learning controller as an intermediary that mediates between alignment parameters and overlay correction. This intermediary layer processes both alignment data and pattern measurement data, separating and reconciling the different parameter types to generate comprehensive correction values that address both substrate-specific and systematic errors.
3Manufacturing precision
If the machine learning controller is trained on measured pattern properties to correct system drift, then system drift correction accuracy is improved, but the device complexity increases
Solution Approach 1:
The patent implements a universal machine learning controller that performs multiple functions: predicting overlay from alignment parameters, correcting substrate-specific variations, and compensating for systematic drift. This multi-functional controller consolidates what would otherwise require separate correction mechanisms, improving precision while managing complexity through a single integrated system that adapts to different correction needs.
Data Source
AI summary
A lithography system configured to apply a pattern to a substrate, the system including a lithography apparatus configured to expose a layer of the substrate according to the pattern, and a machine learning controller configured to control the lithography system to optimize a property of the pattern, the machine learning controller configured to be trained on the basis of a property measured by a metrology unit configured to measure the property of the exposed pattern in the layer and/or a property associated with exposing the pattern onto the substrate, and to correct lithography system drift by adjusting one or more selected from: the lithography apparatus, a track unit configured to apply the layer on the substrate for lithographic exposure, and/or a control unit configured to control an automatic substrate flow among the track unit, the lithography apparatus, and the metrology unit.