Semiconductor Layout Novelty Detection for SEM Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing prediction systems fail to accurately predict SEM patterns for novel input layouts that have not been learned by the model, leading to incorrect pattern generation and reduced accuracy.
Innovation Solution
Incorporating a novelty detection model that uses an auto encoder comparison model and a support vector machine to identify novel layout patterns, allowing only those patterns to be processed for real SEM pattern collection and model retraining, thereby improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the prediction model processes all input layout patterns, then prediction coverage is maintained, but prediction accuracy deteriorates for novel patterns that have not been learned
Solution Approach 1:
A novelty detection model is introduced as an intermediary component between the input layout pattern and the prediction model. This intermediary assesses whether the input pattern is novel or familiar, and only allows familiar patterns to be processed by the prediction model, thereby preventing inaccurate predictions for novel patterns while maintaining efficient processing for known patterns.
Solution Approach 2:
The novelty detection model performs preliminary assessment of input patterns before they are processed by the prediction model. By pre-evaluating whether a pattern is novel or familiar, the system can prepare appropriate handling strategies in advance, ensuring that only suitable patterns are passed to the prediction model for processing.
2Manufacturing precision
If real SEM patterns are collected for all layout patterns, then model training data is comprehensive, but time and resources are wasted on patterns already learned by the model
Solution Approach 1:
The novelty detection model extracts and identifies only the novel layout patterns from the input set, separating them from familiar patterns. This extraction process enables the system to focus resources exclusively on collecting real SEM patterns for novel patterns that require learning, while skipping already-learned patterns, thereby reducing time and resource consumption.
Solution Approach 2:
The system changes the parameter of pattern selection from processing all patterns to processing only novel patterns. By dynamically adjusting which patterns require real SEM pattern collection based on novelty assessment, the system optimizes the balance between training data quality and processing efficiency.
3Productivity
If the system processes only familiar patterns, then processing efficiency is improved, but the system cannot adapt to new pattern types
Solution Approach 1:
The novelty detection model provides feedback about the nature of input patterns (novel vs. familiar) to the overall system. This feedback mechanism enables the system to adapt its processing strategy dynamically: familiar patterns are processed efficiently by the prediction model, while novel patterns trigger data collection and model retraining, ensuring continuous learning and adaptation.
Data Source
AI summary
The invention provides an operation method of a semiconductor system, which includes providing a system which includes a layout pattern to scanning electron microscope (SEM) pattern prediction model (LS model) and a novelty detection model (ND model), inputting a layout pattern to the ND model, and the ND model judges whether the layout pattern is a novel layout pattern, and if the layout pattern is confirmed as the novel layout pattern after judgment, performing a process step on the novel layout pattern to form an SEM (scanning electron microscope) pattern.