Semiconductor Circuit Model Generation via Machine Learning Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As semiconductors become highly integrated and micronized, complex unintended electrical properties occur, necessitating improved methods for predicting and simulating semiconductor device properties to overcome process limitations and reduce experimental costs.
Innovation Solution
The method involves generating circuit models with improved consistency by classifying feature data of a target semiconductor device into first and second feature element data, preprocessing these data sets, extracting machine learning models, and combining them to create a circuit model for simulating integrated circuits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional circuit modeling methods are used for highly integrated and micronized semiconductors, then the design and manufacturing process can proceed with conventional tools, but the accuracy of predicting electrical properties deteriorates due to complex unintended electrical properties
Solution Approach 1:
The patent segments the complex semiconductor device modeling into multiple independent feature elements (e.g., gate capacitance, drain capacitance, source capacitance, resistance elements). Each feature element is modeled separately using machine learning, allowing the system to handle complex unintended electrical properties through modular, independent predictions that are then combined to form the complete circuit model.
2Measurement precision
If more experiments are conducted to understand semiconductor phenomena and overcome process limitations, then the accuracy of device property prediction improves, but the cost and time of experiments increase
Solution Approach 1:
The patent creates virtual copies of semiconductor device behavior through machine learning models trained on experimental data. These digital twins (circuit models) replicate the electrical properties and responses of physical devices, enabling accurate prediction and analysis without requiring additional physical experiments. The models can simulate various conditions and scenarios that would be costly or time-consuming to test physically.
3Reliability
If circuit models are generated without classifying feature data by measurement conditions, then the modeling process is simpler and faster, but the consistency and accuracy of the circuit model deteriorates
Solution Approach 1:
The patent applies local quality by tailoring the modeling approach to specific measurement conditions. Feature data is classified according to different measurement conditions (e.g., frequency ranges, bias conditions, temperature), and separate machine learning models are trained for each condition. This ensures that each local aspect of the device behavior is accurately captured under its specific operating conditions, improving overall model consistency.
Solution Approach 2:
The patent systematically varies and analyzes different parameters (measurement conditions, feature elements, model architectures) to optimize the circuit model. By changing parameters such as the classification criteria for feature data and the specific machine learning algorithms used, the system adapts to capture the nuanced behavior of semiconductor devices under different operating conditions, thereby improving reliability.
Data Source
AI summary
A method of generating a circuit model used to simulate an integrated circuit may include generating first feature element data and second feature element data by classifying feature data of a target semiconductor device according to measurement conditions, generating first target data and second target data by preprocessing the first feature element data and the second feature element data, respectively, generating a first machine learning model using the first target data and extracting a second machine learning model using the second target data, and generating the circuit model used to simulate the integrated circuit using the first machine learning model and the second machine learning model.