Neural Network Modeling for Semiconductor Simulation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity and miniaturization of semiconductors lead to complex interactions in design and manufacturing, making it challenging to predict and simulate semiconductor device characteristics accurately, which is costly and requires advanced simulation technologies.
Innovation Solution
A neural network modeling method that trains and re-trains regression models based on sample data and simulation results, using active and balanced sampling to improve consistency and automate the training process, thereby enhancing prediction accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If advanced simulation technologies are used to predict semiconductor device characteristics, then prediction accuracy is improved, but data collection time and cost increase
Solution Approach 1:
The patent applies preliminary action by performing active sampling to identify and prioritize regions with prediction failures before conducting full-scale simulations. This allows the system to pre-determine which areas need detailed analysis, reducing overall data collection time while maintaining prediction accuracy in critical regions
Solution Approach 2:
The patent implements local quality by applying different sampling strategies to different regions of the semiconductor device characteristics space. Balanced sampling is used in regions where prediction consistency is insufficient, while other regions use standard sampling. This localized approach improves prediction accuracy where needed without uniformly increasing data collection time across all regions
2Reliability
If comprehensive training data is collected to improve model consistency, then model reliability is improved, but training complexity and resource requirements increase
Solution Approach 1:
The patent applies feedback by using the first regression model to identify prediction failures, then using this feedback information to guide active sampling for generating second training data. The system continuously monitors model performance and directs additional training data collection to specific regions where consistency is insufficient, improving model reliability without requiring comprehensive re-training across all parameters
Solution Approach 2:
The patent implements segmentation by dividing the training process into two distinct phases: initial training with first sample data to establish baseline performance, and targeted re-training with second sample data focused on prediction failure regions. This segmented approach reduces training complexity by avoiding uniform re-training of the entire model and instead focusing computational resources on specific problematic areas
3Measurement precision
If manual engineer intervention is used to improve model performance, then prediction accuracy is improved, but automation level decreases
Solution Approach 1:
The patent applies self-service by enabling the regression model to automatically identify its own prediction failures and guide the collection of additional training data without engineer intervention. The system autonomously determines which regions need improved prediction accuracy and automatically generates targeted training data, maintaining high prediction accuracy while achieving full automation
Solution Approach 2:
The patent uses feedback mechanisms where the model's prediction results automatically feed into the sampling strategy for generating additional training data. This closed-loop system eliminates the need for manual engineer intervention by allowing the system to self-correct prediction failures through automated identification and targeted re-training
Data Source
AI summary
In a modeling method of a neural network, a first regression model is trained based on first sample data and first simulation result data. The first regression model is used to predict the first simulation result data from the first sample data. The first sample data represent at least one of conditions of a manufacturing process of a semiconductor device and characteristics of the semiconductor device. The first simulation result data are obtained by performing a simulation on the first sample data. In response to a consistency of the first regression model being lower than a target consistency, the first regression model is re-trained based on second sample data different from the first sample data. The second sample data are associated with a consistency reduction factor of the first regression model that is responsible for a prediction failure of the first regression model.