Semiconductor Device Characteristic Prediction Using TCAD Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing design and simulation processes for semiconductor products are inefficient and costly due to the need for frequent changes in device models when process levels change, leading to inaccurate predictions and performance degradation.
Innovation Solution
A method using deep learning to predict semiconductor device characteristics by generating training data from TCAD simulator results and training a deep learning model to output accurate prediction data, with uncertainty handling and retraining based on actual devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a TCAD simulator is used to generate device models for semiconductor design, then simulation accuracy is improved, but development time and cost increase due to frequent model changes when process levels change
Solution Approach 1:
The patent creates a simplified copy of the complex TCAD simulator through a deep learning model. This neural network model learns the input-output relationships from TCAD simulation data and provides rapid predictions without executing the full physical simulation, thus maintaining accuracy while dramatically reducing computation time
Solution Approach 2:
The system performs preliminary training of the deep learning model using historical TCAD simulation data before actual design work begins. This pre-trained model can then quickly predict device characteristics for new process levels without requiring time-consuming re-simulation, enabling faster design iterations
2Measurement precision
If frequent simulation re-performance is conducted when process levels change, then prediction accuracy is maintained, but manufacturing cost increases
Solution Approach 1:
The deep learning model serves as a computational copy that encapsulates the essential physics and design rules learned from TCAD simulations. Once trained, it can predict device characteristics for various process levels without requiring expensive and time-consuming re-execution of the full simulation suite, thereby reducing manufacturing costs while maintaining prediction accuracy
Solution Approach 2:
The system adapts to process level changes by adjusting input parameters to the pre-trained deep learning model rather than re-running simulations. The model generalizes across different process levels through its training on diverse data, allowing accurate predictions for new process nodes without costly re-simulation
3Loss of time
If low accuracy simulation models are used, then development time is reduced, but product performance degrades due to inaccurate characteristics prediction
Solution Approach 1:
The patent creates an accurate copy of the TCAD simulator's predictive capability through deep learning. The neural network model is trained to reproduce the same level of accuracy as the original simulator while operating much faster, thus avoiding the trade-off between speed and accuracy that plagues conventional simulation approaches
Solution Approach 2:
The system incorporates feedback mechanisms where prediction results are compared with actual measured data from manufactured devices. This feedback loop enables continuous refinement and retraining of the deep learning model, ensuring that predictions remain accurate as process levels evolve and product performance requirements change
Data Source
AI summary
To predict characteristics of a semiconductor device, basic training data corresponding to a combination of input data and simulation result data are generated using a technology computer aided design (TCAD) simulator. The TCAD simulator generates the simulation result data by performing a simulation based on the input data of the TCAD simulator such that the simulation result data indicates characteristics of semiconductor devices corresponding to the input data of the TCAD simulator. A deep learning model is trained based on the basic training data such that the deep learning model is configured to output prediction data indicating the characteristics of the semiconductor devices. Target prediction data is generated based on the deep learning model and input data corresponding to the target semiconductor product such that the target prediction data indicates the characteristics of the semiconductor device included in the target semiconductor product.