Neural Network Model for Semiconductor Circuit Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The semiconductor industry faces challenges in predicting and simulating the complex electrical characteristics of miniaturized semiconductor devices, which are exacerbated by the intricate interactions of design and manufacturing stages, leading to unintended electrical characteristics and increased experimental costs.
Innovation Solution
A method and computing device for generating a neural network model with improved performance and reduced cost, and performing circuit simulations using this model. The approach involves generating sample data through process simulations based on temperature and process parameters, training the neural network model, performing lightweight operations to reduce complexity, re-training the model, and using it for circuit simulations with process parameters as input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional process simulation is used for semiconductor device characterization, then measurement precision is improved, but loss of time increases significantly
Solution Approach 1:
The patent performs process simulation in advance to generate training data for a neural network model. The neural network is trained beforehand with simulation data covering various process parameters and temperature conditions. During actual device characterization, the pre-trained neural network rapidly predicts electrical characteristics without requiring real-time process simulation, thus achieving both high precision and fast performance.
Solution Approach 2:
The patent creates a neural network model that copies the complex physical relationships learned from detailed process simulations. Instead of running the original complex simulation repeatedly, the neural network serves as a simplified copy that reproduces simulation results instantaneously. This copying approach maintains measurement precision while dramatically reducing the time required for device characterization.
2Reliability
If detailed process simulation is performed to understand semiconductor phenomena, then reliability is improved, but productivity decreases due to computational complexity
Solution Approach 1:
The patent performs comprehensive process simulations in advance to train the neural network model with accurate physical data. Once trained, the neural network can rapidly perform device characterization and parameter extraction for multiple process conditions without repeating the full simulation. This preliminary action ensures high reliability through thorough initial training while improving productivity during subsequent optimization iterations.
Solution Approach 2:
The neural network acts as an intermediary between the complex process simulation and the device characterization task. Instead of directly running expensive simulations for every parameter extraction, the neural network mediates by providing rapid predictions based on its pre-learned physical relationships. This intermediary approach maintains the reliability of physics-based modeling while enabling fast productivity for design optimization.
3Measurement precision
If neural network model complexity is increased to improve prediction accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent employs data normalization techniques to transform input parameters (such as scaling voltage, current, and dimensional parameters) into standardized ranges. This parameter transformation allows the neural network to achieve high prediction accuracy with a simpler architecture, as normalized data converges faster during training and requires fewer hidden layers or neurons. The normalization itself captures complex physical relationships without increasing the explicit model complexity.
Solution Approach 2:
The patent applies selective data augmentation by generating training samples only for critical process parameters and temperature ranges that most significantly impact device characteristics. Rather than comprehensively simulating all possible parameter combinations, the method focuses on the most influential factors, achieving high measurement precision for practical device characterization while keeping the training dataset and model complexity manageable.
Data Source
AI summary
A method of generating a neural network model and performing a circuit simulation by using the neural network model is presented. The method includes generating sample data by performing a process simulation based on a temperature and a process parameter, training the neural network model based on the sample data, performing a lightweight operation on the neural network model to generate a lightweight neural network model, re-training the lightweight neural network model, and performing the circuit simulation with the process parameter as an input by using the re-trained lightweight neural network model.