Charged-Particle Beam Drift Prediction for Photolithography
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In photolithography, the integration of smaller structures on semiconductor wafers leads to complex and costly mask production, with errors due to electrostatic charging and thermal drift causing beam distortions, making precise drift compensation challenging.
Innovation Solution
A device using a trained machine learning model or predictive filter to continuously predict and compensate for the drift of a charged particle beam relative to a photolithographic element, based on measurement data and additional parameters, allowing for reliable and efficient drift correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If reference markings are scanned frequently to detect displacement accurately, then drift compensation precision is improved, but marking degradation increases and measurement time increases
Solution Approach 1:
The patent applies preliminary action by training a machine learning model offline using historical measurement data before actual drift compensation. The model learns to predict drift based on process parameters (temperature, pressure, gas flow rates) without requiring frequent real-time measurements of reference markings. This allows the system to make accurate drift predictions during operation without the time penalty of frequent physical measurements.
Solution Approach 2:
The patent replaces the mechanical measurement system (physically scanning reference markings with a beam) with a computational model. Instead of repeatedly measuring the position of reference markings, the system uses a trained machine learning model that processes easily measurable process parameters to predict drift. This substitution dramatically reduces measurement time while maintaining or improving prediction accuracy.
2Measurement precision
If reference markings are scanned frequently to detect displacement, then drift compensation accuracy is improved, but marking degradation increases
Solution Approach 1:
The patent substitutes the physical measurement of reference markings with a computational prediction system. The machine learning model predicts drift based on process parameters without requiring physical interaction with reference markings during operation. This eliminates the degradation problem entirely while maintaining drift detection accuracy through the learned relationships in the model.
Solution Approach 2:
The patent introduces process parameters (temperature, pressure, gas flow rates) as intermediary variables that correlate with drift but do not require direct measurement of reference markings. These intermediaries serve as proxies that can be measured without degrading the sample, and the machine learning model learns the relationship between these intermediaries and actual drift.
3Reliability
If traditional drift compensation methods are used with frequent measurements, then drift correction reliability is improved, but device complexity and processing time increase
Solution Approach 1:
The patent replaces time-consuming physical measurement and correction cycles with a computational model that processes quickly. The machine learning model can evaluate drift based on process parameters in real-time without the overhead of acquiring, processing, and analyzing images of reference markings. This maintains drift correction reliability while significantly improving processing throughput.
Solution Approach 2:
The patent performs the complex work of learning drift patterns during offline training before production use. During actual photolithography operations, the pre-trained model requires only simple parameter input to predict drift, eliminating the need for complex real-time image processing and analysis. This preliminary preparation enables fast, reliable drift correction during high-speed production.
Data Source
AI summary
A device for examining and/or processing an element for photolithography with a beam of charged particles, the device including (a) means for acquiring measurement data while the element for photolithography is exposed to the beam of charged particles; and (b) means for predetermining a drift of the beam of charged particles relative to the element for photolithography with a trained machine learning model and/or a predictive filter. The trained machine learning model and/or the predictive filter use(s) at least the measurement data as input data.