Patterning Process Control Using ML Models for Drift Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing lithographic systems face challenges in accurately modeling motion, disturbance forces, heating, and other patterning process conditions that change over time, leading to inaccuracies in feature reproduction and process control.
Innovation Solution
Implementing a trained machine learning model to generate control outputs for patterning processes, using training data from simulations and actual process data to adjust parameters such as motion, heating, and disturbance forces, and updating the model over time to account for drift and changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical models are used to model motion, disturbance forces, heating, and other patterning process conditions, then the modeling accuracy is improved, but the computational expense increases and the models require frequent adjustment when process conditions change
Solution Approach 1:
The patent creates a virtual copy of the physical patterning process through simulation models that replicate the behavior of motion, heating, and disturbance forces. These virtual models allow accurate prediction of process outcomes without requiring expensive physical experimentation or complex real-time physical modeling, thereby reducing computational expense while maintaining modeling accuracy.
Solution Approach 2:
The patent performs simulations and generates training data in advance before actual patterning processes occur. By pre-computing the effects of various process conditions through physics-based simulations, the system establishes a library of expected outcomes that can be quickly referenced during production, eliminating the need for expensive real-time physical modeling and reducing computational expenses during actual manufacturing.
2Measurement precision
If physical models are adjusted to account for changes in process conditions, then the modeling accuracy is improved, but the time required for model calibration and setup increases
Solution Approach 1:
The patent uses virtual simulations to create representative models of various process conditions that can be stored and quickly applied. Instead of recalibrating physical models for each new process condition, the system selects or interpolates from pre-computed virtual models that match the current conditions, dramatically reducing calibration time while maintaining accuracy.
Solution Approach 2:
The patent performs extensive model calibration and adjustment work in advance through simulation studies. By pre-establishing the relationships between process parameters and outcomes across a wide range of conditions, the system eliminates the need for time-consuming real-time calibration, as the virtual models are already optimized for various scenarios before production begins.
3Manufacturing precision
If frequent measurements and adjustments are made to maintain pattern reproduction accuracy, then the manufacturing precision is improved, but the productivity decreases
Solution Approach 1:
The patent replaces physical measurement and adjustment operations with virtual sensing and control through simulation models. By using computational models to predict process outcomes and guide adjustments, the system eliminates the need for frequent physical measurements and manual interventions, thereby maintaining high manufacturing precision while improving productivity through faster, automated decision-making.
Solution Approach 2:
The patent performs virtual measurements and optimizations in advance through simulation. By predicting process outcomes and identifying optimal parameters before actual production, the system reduces the need for frequent real-time measurements and adjustments during manufacturing, thus maintaining precision while maximizing throughput.
Data Source
AI summary
Generating a control output for a patterning process is described. A control input is received. The control input is for controlling the patterning process. The control input includes one or more parameters used in the patterning process. The control output is generated with a trained machine learning model based on the control input. The machine learning model is trained with training data generated from simulation of the patterning process and/or actual process data. The training data includes 1) a plurality of training control inputs corresponding to a plurality of operational conditions of the patterning process, where the plurality of operational conditions of the patterning process are associated with operational condition specific behavior of the patterning process over time, and 2) training control outputs generated using a physical model based on the training control inputs.