Transfer Learning Simulation Models for Semiconductor Process Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for improving semiconductor process simulation accuracy by adjusting parameters based on physical knowledge or applying neural networks are insufficient in reducing the difference between simulation data and real measurement data.
Innovation Solution
A method involving a neural network device that classifies and processes weight data by separating weight parameters into groups based on significance, retraining the first weight group using simulation data, and training the second weight group using measurement data within a transfer learning model to generate a simulation model that aligns with measurement data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural network technology is applied to improve simulation performance, then the simulation accuracy can be enhanced, but the difference between simulation data and real measurement data remains insufficiently reduced
Solution Approach 1:
The patent segments weight parameters into two distinct groups: first weight parameters that are retrained using simulation data, and second weight parameters that are trained using measurement data. This segmentation allows the model to selectively incorporate information from both simulation and measurement data, thereby reducing the difference between simulation results and real measurements while maintaining simulation accuracy.
Solution Approach 2:
The patent applies local quality by treating different weight parameters with different training strategies. Specifically, certain weight parameters are optimized using simulation data while others are optimized using measurement data. This localized differentiation enables the model to capture both the general patterns from simulation and the specific characteristics from measurements, improving the alignment between simulation and reality.
2Measurement precision
If parameter calibration is performed by directly adjusting parameters based on physical knowledge, then simulation performance can be improved, but the reduction in difference between simulation and measurement data is insufficient
Solution Approach 1:
The patent introduces a transfer learning model as an intermediary that bridges simulation data and measurement data. This model learns from both data sources and generates corrected simulation results that align with measurement data. The transfer learning model acts as a mediator that translates physical knowledge from simulation into predictions that match real-world measurements, overcoming the limitations of direct parameter calibration.
Data Source
AI summary
A method of generating a simulation model based on simulation data and measurement data of a target includes classifying weight parameters, included in a pre-learning model learned based on the simulation data, as a first weight group and a second weight group based on a degree of significance, retraining the first weight group of the pre-learning model based on the simulation data, and training the second weight group of a transfer learning model based on the measurement data, wherein the transfer learning model includes the first weight group of the pre-learning model retrained based on the simulation data.