Semiconductor Structure Prediction via Meta-Learning Stacking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor device structure prediction methods face challenges with overfitting due to insufficient training data and require extensive sample preparation, leading to inefficiencies in simulator performance and accuracy.
Innovation Solution
A device structure simulation apparatus and method using a stacking model and meta-learning model to predict semiconductor device structures, which preprocesses spectrum data and combines simulation and measurement data from multiple sub-models to enhance prediction accuracy and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural network technology is used to predict device structure, then prediction accuracy is improved, but overfitting occurs due to insufficient training data
Solution Approach 1:
The patent divides the training data into multiple subsets (simulation data and measurement data) and trains separate sub-models for each subset. This segmentation allows each sub-model to specialize in specific data characteristics, reducing overfitting while maintaining overall prediction accuracy through the ensemble of specialized models.
Solution Approach 2:
The patent combines multiple sub-models (first sub-model trained on simulation data, second sub-model trained on measurement data) into a unified prediction system. This merging leverages the strengths of different data sources and models, improving robustness and reducing overfitting vulnerability through ensemble methods.
2Reliability
If extensive sample preparation and reference data measurement are performed, then prediction reliability is improved, but time and cost increase
Solution Approach 1:
The patent performs preliminary action by pre-training a meta-model using simulation data before actual measurement. This pre-training establishes a foundational understanding of device structures that can be quickly adapted to new measurement data, reducing the time needed for sample preparation and reference data measurement while maintaining prediction reliability.
Solution Approach 2:
The patent uses simulation data as a copy or virtual representation of actual device structures to pre-train the meta-model. This copying allows extensive training on diverse structures without requiring physical sample preparation, significantly reducing time and cost while maintaining the ability to accurately predict real device structures.
3Measurement precision
If engineers personally adjust parameters using physical knowledge, then prediction accuracy is improved, but device complexity and operational burden increase
Solution Approach 1:
The patent implements self-service by enabling the system to automatically adjust parameters through the meta-learning framework. The meta-model automatically adapts to new data and adjusts its internal parameters without requiring manual intervention from engineers, maintaining high prediction accuracy while reducing operational complexity and burden.
Data Source
AI summary
A device structure simulation apparatus includes a memory storing a device structure simulation program and a processor configured to execute the device structure simulation program stored in the memory. By executing the device structure simulation program, the device structure simulation apparatus is further configured to receive spectrum data of a target device, generate an input data set by performing preprocessing on the spectrum data, and train a model based on the input data set such that the model is configured to predict a structure of the target device. The preprocessing including selecting a certain basis function based on the spectrum data and separating the spectrum data into sets of certain basis functions, and the model includes at least one sub model.