Semiconductor Power Module Current Balancing Using Reinforcement Machine Learning

In high power applications, semiconductor power modules containing paralleled MOSFETs are often used to achieve high output currents. The current distribution between devices within a module is influenced by several factors such as component layout, minor defects due to manufacturing tolerances, and...

Full description

Saved in:
Bibliographic Details
Published in2021 IEEE Pulsed Power Conference (PPC) pp. 1 - 5
Main Authors Westmoreland, B., Bilbao, A. V., Bayne, S. B.
Format Conference Proceeding
LanguageEnglish
Published IEEE 12.12.2021
Subjects
Layout
MOSFET
Multichip modules
Reinforcement learning
Semiconductor device measurement
Software packages
Voltage measurement
Online AccessGet full text

Cover

More Information
Summary:In high power applications, semiconductor power modules containing paralleled MOSFETs are often used to achieve high output currents. The current distribution between devices within a module is influenced by several factors such as component layout, minor defects due to manufacturing tolerances, and general devices degradation that occurs over time. This paper describes a method of balancing the current between paralleled MOSFETs by independently modulating each device's gate-to-source voltage and measuring the corresponding drain-to-source currents. To achieve this, a detailed simulation is created using MATLAB and Simulink. A reinforcement learning agent is implemented with the goal of adaptively balancing power module current as the components inside degrade over time.
ISSN:2158-4923
DOI:10.1109/PPC40517.2021.9733124