MOSFET RDS-ON Prediction via Transient Current and Voltage Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting the drain-to-source resistance (RDS-ON) of power metal-oxide-semiconductor field-effect transistors (MOSFETs) lack flexibility, accuracy, and resilience against noise, leading to inadequate predictive capabilities and increased risk of device failure.
Innovation Solution
A system utilizing a current sensing circuit, conditioning circuit, and machine learning model to monitor and predict future RDS-ON values by capturing drain current and voltage during transient phases, employing a Levenberg-Marquardt algorithm for noise mitigation and regularization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods are used to predict RDS-ON values, then the system is simpler to implement, but predictive accuracy and reliability deteriorate
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between the measured electrical parameters (drain current, drain-to-source voltage) and the predicted RDS-ON values. This ML intermediary processes the input signals through trained algorithms to produce accurate predictions, resolving the contradiction by providing high predictive accuracy while maintaining a manageable system structure through modular architecture.
Solution Approach 2:
The patent replaces conventional empirical or physics-based prediction methods with a data-driven machine learning approach. By substituting traditional mechanical or analytical calculation systems with ML models trained on operational data, the system achieves superior predictive accuracy without proportionally increasing physical complexity.
2Reliability
If conventional prediction methods are used, then the device complexity is lower, but reliability and resilience against noise worsen
Solution Approach 1:
The patent implements feedback mechanisms where the machine learning model continuously receives operational data from the power MOSFET, processes it, and updates predictions. The system incorporates feedback loops that allow the model to adapt to changing conditions and noise patterns, improving reliability through continuous learning and adjustment while maintaining system manageability.
Solution Approach 2:
The patent applies preliminary action by training the machine learning model in advance using extensive datasets that include noisy operational conditions. This pre-training prepares the model to handle noise and variability during actual operation, enhancing reliability before the system is deployed without adding complexity during runtime.
3Measurement precision
If machine learning models are implemented for prediction, then predictive accuracy improves, but computational requirements and processing time increase
Solution Approach 1:
The patent performs the computationally intensive model training in advance, before actual prediction operations. By completing the heavy computational work during the training phase using extensive datasets, the system achieves high predictive accuracy while keeping runtime processing requirements minimal, thus resolving the contradiction between accuracy and processing time.
Solution Approach 2:
The patent uses simplified copies or approximations of the full machine learning model for real-time predictions. By deploying optimized versions of the trained model (such as quantized weights, pruned networks, or surrogate models), the system maintains high predictive accuracy while significantly reducing computational burden and processing time during operation.
Data Source
AI summary
According to an embodiment, a system is proposed for predicting future drain-to-source resistance values of a power metal-oxide-semiconductor field-effect transistor (MOSFET). The system includes a current sensing circuit configured to extract a drain current of the power MOSFET during a transient or switching phase of the power MOSFET; a conditioning circuit configured to extract a drain-to-source voltage of the power MOSFET during the transient or switching phase of the power MOSFET; and a machine learning model configured to receive the drain current and the drain-to-source voltage of the power MOSFET as inputs, and determine future drain-to-source resistance values of the power MOSFET using the machine learning model based on the extracted drain current and the drain-to-source voltage of the power MOSFET, wherein a fault condition of the power MOSFET is signaled in response to the predicted future drain-to-source resistance values exceeding a threshold.