Operator-Learned Semiconductor Simulation Beyond Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing TCAD simulators face challenges in accurately predicting semiconductor characteristics due to the increasing complexity of semiconductor manufacturing processes, leading to difficulties in accurately simulating semiconductor devices.
Innovation Solution
A simulation method and system utilizing a machine learning model, specifically an artificial neural network pre-trained through operator learning, directly learns the solution of the drift-diffusion equation, enabling accurate prediction of semiconductor device attributes even outside the trained data range.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional TCAD simulators are used to predict semiconductor characteristics, then simulation coverage can be maintained, but prediction accuracy deteriorates as manufacturing process complexity increases
Solution Approach 1:
The patent replaces the traditional TCAD simulator (mechanical/computational system) with a machine learning model (data-driven system). The ML model learns patterns from training data and directly predicts semiconductor characteristics without requiring complex physical simulations, thereby maintaining or improving accuracy while reducing the burden of process complexity.
Solution Approach 2:
The patent transforms the simulation approach by changing from physical parameter-based TCAD simulation to data-parameter-based machine learning prediction. The ML model uses learned parameters from training data to predict characteristics, avoiding the need to explicitly model complex manufacturing processes.
2Measurement precision
If machine learning models are trained on extensive data to improve accuracy, then prediction performance improves, but memory and computational requirements increase
Solution Approach 1:
The patent applies partial training action by training the ML model on a selective subset of training data rather than exhaustively training on all available data. This partial training approach achieves sufficient prediction accuracy while significantly reducing memory and computational resource requirements compared to comprehensive training.
Solution Approach 2:
The patent performs preliminary selection of training data to identify the most relevant and informative samples. By pre-selecting key training examples that capture essential patterns, the model achieves good prediction performance with reduced training data volume, thereby lowering computational and memory requirements.
Data Source
AI summary
A method of performing a process simulation of a semiconductor device may include obtaining a target parameter and first state profile data corresponding to an initial value; and generating second state profile data corresponding to the target parameter from the first state profile data, based on a machine learning model. Each of the first state profile data and the second state profile data may represent an attribute profile of a corresponding state of a semiconductor device.