Photoresist Model Calibration with Bayesian Active Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing photoresist models in photolithography processes lack accuracy and require extensive manual parameter adjustments due to the introduction of new photographing technologies and photoresists, making it difficult to optimize parameters effectively.
Innovation Solution
A method and apparatus for determining a photoresist model that iteratively fits and optimizes parameters using Gaussian process models and Bayesian active learning to predict photoresist patterns, allowing for collaborative optimization of coefficients and parameters, reducing the need for manual adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional photoresist models are used with new photographing technologies and photoresists, then the manufacturing process can continue, but the model accuracy deteriorates and extensive manual parameter adjustments are required
Solution Approach 1:
The patent applies parameter changes by systematically adjusting photoresist model parameters through automated algorithms. The system modifies parameters such as exposure dose, development time, and chemical concentrations based on process conditions and measurement data, replacing manual parameter tuning with computational optimization to maintain model accuracy under new photographing technologies
Solution Approach 2:
The patent implements feedback mechanisms by measuring actual photoresist pattern results and using this information to iteratively refine model parameters. The system continuously compares predicted patterns with measured patterns, adjusting parameters based on the differences, thereby maintaining high model accuracy without extensive manual intervention
2Measurement precision
If extensive manual parameter adjustments are performed to improve photoresist model accuracy, then model precision improves, but the time consumption and optimization efficiency deteriorate
Solution Approach 1:
The patent applies self-service by enabling the photoresist model to automatically adjust its own parameters through embedded optimization algorithms. The system performs self-calibration by comparing predicted patterns with actual measurements and autonomously refining parameters, eliminating the need for time-consuming manual optimization while maintaining high prediction accuracy
Solution Approach 2:
The patent implements preliminary action by pre-configuring optimization algorithms and parameter ranges before the actual modeling process. The system prepares lookup tables, establishes initial parameter bounds, and sets up automated adjustment rules in advance, enabling rapid parameter optimization without time-consuming manual trial and error during production
3Reliability
If more parameter groups are included in the photoresist model to improve accuracy, then the model becomes more comprehensive, but the calculation complexity and optimization difficulty increase
Solution Approach 1:
The patent applies segmentation by dividing the comprehensive parameter space into distinct parameter groups that can be optimized independently. The system separates parameters into categories such as exposure parameters, development parameters, and material-specific parameters, allowing targeted optimization of each group without overwhelming computational complexity while maintaining overall model comprehensiveness
Data Source
AI summary
A method and an apparatus for determining a photoresist model used for generating a photoresist pattern. The method includes: obtaining a plurality of first parameter groups from a first parameter group set of a photoresist model; fitting the photoresist model based on the plurality of first parameter groups and a plurality of reference photoresist pattern matrixes to obtain a plurality of coefficients; predicting expected values of a plurality of second parameter groups based on the photoresist model with the determined coefficients and errors between predicted photoresist pattern matrixes of first parameter groups and the reference photoresist pattern matrixes, and adding a second parameter group with the expected value to obtain an updated first parameter group set; and respectively assigning a plurality of candidate values in the parameter group having a minimum error in the updated first parameter group set to a plurality of parameters in the photoresist model.