Deep Learning Model for Semiconductor Device Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the semiconductor foundry industry, changing a semiconductor product design during simulation can be time-consuming and costly, and inaccurate prediction models can lead to degraded performance of the manufactured product due to low accuracy in predicting device characteristics.
Innovation Solution
A method of generating a deep learning model that includes generating basic training data from device and simulation data using a compact model, training the model to output prediction and uncertainty data, and retraining the model based on uncertainty data to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simulation is performed based on changed design, then prediction accuracy is improved, but time loss and cost increase enormously
Solution Approach 1:
The patent trains a deep learning model in advance using comprehensive simulation data covering various design conditions. This pre-trained model can then quickly predict characteristics for new designs without requiring time-consuming re-simulation, thus resolving the contradiction between prediction accuracy and time loss.
Solution Approach 2:
The patent creates a deep learning model that copies the behavior and characteristics of complex simulation results. Instead of performing full simulations for each design change, the model replicates simulation outcomes based on input parameters, providing accurate predictions without the time cost of actual simulations.
2Measurement precision
If simulation is performed based on changed design, then prediction accuracy is improved, but extra cost increases
Solution Approach 1:
The deep learning model is trained beforehand using simulation data, so that when design changes occur, predictions can be made instantly without incurring additional simulation costs. This preliminary preparation eliminates the need for expensive re-simulations while maintaining accuracy.
Solution Approach 2:
The model copies simulation results and uses them for prediction, avoiding the need to perform expensive simulations repeatedly. This copying approach maintains prediction accuracy while eliminating the recurring cost burden.
3Device complexity
If deep learning model outputs only prediction data, then model complexity is reduced, but reliability decreases due to inability to assess prediction confidence
Solution Approach 1:
The patent segments the model output into distinct components: prediction data and uncertainty data. This segmentation allows the model to provide both the predicted characteristics and a measure of confidence, improving reliability without significantly increasing overall model complexity.
Solution Approach 2:
The patent adds uncertainty data as an additional output dimension alongside prediction data. This extra dimension provides information about prediction confidence, enhancing reliability while maintaining a relatively simple model structure that outputs multiple values.
Data Source
AI summary
To generate a deep learning model, basic training data corresponding to a combination of device data and simulation result data is generated using a compact model that generates the simulation result data indicating characteristics of a semiconductor device corresponding to the device data by performing simulation based on the device data. A deep learning model is trained based on the basic training data such that the deep learning model outputs prediction data indicating the characteristics of the semiconductor device and uncertainty data indicating uncertainty of the prediction data. The deep learning model is retrained based on the uncertainty data. The deep learning model may precisely predict the characteristics of the semiconductor device by training the deep learning model to output the prediction data and the uncertainty data and retraining the deep learning model based on the uncertainty data.