Encoder-Decoder Model Uncertainty Quantification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models in photolithography lack clarity in certainty of predictions, leading to uncertainties in mask layout generation and increased uncertainty in integrated circuit manufacturing processes, with no existing method to determine variability in model output.
Innovation Solution
A method using an encoder-decoder architecture with variational inference techniques to generate distributions of posterior probabilities, sample from these distributions, and quantify uncertainty by determining variability in predicted outputs, allowing for adjustment of model parameters to reduce uncertainty by increasing training data and dimensionality of the latent space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning models are used for predictions in photolithography, then productivity is improved, but reliability deteriorates due to uncertainty in predictions
Solution Approach 1:
The patent implements a feedback mechanism by computing uncertainty metrics from the machine learning model predictions and using this information to iteratively refine the models. The uncertainty quantification provides feedback about model confidence, allowing for continuous improvement of prediction reliability while maintaining high productivity.
Solution Approach 2:
The patent replaces traditional deterministic mechanical/mathematical modeling approaches with machine learning models that inherently provide uncertainty quantification. This substitution allows the system to maintain productivity benefits of ML while addressing reliability concerns through probabilistic predictions and uncertainty metrics.
2Reliability
If machine learning models are adjusted to reduce uncertainty, then reliability is improved, but device complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-computing uncertainty metrics and validating model predictions before they are used in critical manufacturing decisions. This allows for risk mitigation without requiring complex model adjustments, maintaining reliability while avoiding unnecessary complexity.
Solution Approach 2:
The patent adjusts model parameters and architecture specifically to optimize uncertainty quantification rather than purely maximizing prediction accuracy. This targeted parameter adjustment reduces model complexity compared to comprehensive model redesign, achieving reliability improvement with minimal complexity increase.
Data Source
AI summary
Described herein is a method for quantifying uncertainty in parameterized (e.g., machine learning) model predictions. The method comprises causing a parameterized model to predict multiple posterior distributions from the parameterized model for a given input. The multiple posterior distributions comprise a distribution of distributions. The method comprises determining a variability of the predicted multiple posterior distributions for the given input by sampling from the distribution of distributions; and using the determined variability in the predicted multiple posterior distributions to quantify uncertainty in the parameterized model predictions. The parameterized model comprises encoder-decoder architecture. The method comprises using the determined variability in the predicted multiple posterior distributions to adjust the parameterized model to decrease the uncertainty of the parameterized model for predicting wafer geometry, overlay, and/or other information as part of a semiconductor manufacturing process.