Real-Time Junction Temperature Estimation for Using Temperature-Sensitive Electrical Parameters

Abstract

Silicon Carbide Metal-Oxide-Semiconductor Field-Effect Transistors (SiC MOSFETs) are core components in high-power systems like new energy vehicle powertrains and solar inverters, thanks to their high breakdown voltage, low on-state loss, and strong high-temperature tolerance. However, their junction temperature (T_J ) rises sharply under high-power/frequency operations, and extreme T_J fluctuations cause solder cracking or wire bonding detachment, shortening service life and risking system failure. Accurate real-time T_J measurement is critical for reliability. Traditional methods have flaws: infrared thermometers lack real-time capability, and built-in sensors require device redesign. Temperature-Sensitive Electrical Parameters (TSEPs)—which use the device’s intrinsic electrical parameters (correlated with T_J )—offer a non-invasive solution without physical contact or design changes, becoming a key research focus. This paper first classifies common TSEPs (static: R_DS,ON, V_DS,ON; dynamic: T_D,OFF, V_VTH) and explains their temperature sensitivity and application scenarios. It then elaborates on three T_J prediction models: traditional mathematical models (polynomial fitting, thermal network fusion), data-driven models (BPNN, DNN), and multi-TSEP fusion models. A specialized experimental circuit for TSEP acquisition (using Kelvin source parasitic inductance and digital isolation) is also presented, enabling real-time measurement in one PWM cycle. Finally, the paper identifies practical challenges (parasitic interference, aging-induced accuracy loss, poor extreme-condition adaptability) and proposes solutions like Kelvin connections and adaptive calibration. Future directions include exploring stable TSEPs and integrating measurement circuits into miniaturized chips, supporting SiC MOSFET thermal management and reliability.