A research team led by Litong Wang, Xiaodong Li, Haifeng Sun, and Xin Liu has developed a three-dimensional semi-empirical closed-form healthy-state baseline model capable of accurately tracking the internal junction temperature of high-power insulated gate bipolar transistors across exceptionally broad operating currents. Documented in the Journal of Engineering and Applied Science, the theoretical framework correlates device on-state voltage drop, collector-emitter current, and junction temperature under varying gate-drive voltages. When validated experimentally against high-current test data, the unified formulation demonstrated an absolute junction temperature estimation error no greater than 2.91 degrees Celsius across a thermal span of 25 to 105 degrees Celsius. Crucially, the model maintained high predictive fidelity even when electrical currents expanded up to approximately 4.5 times the rated device capacity, providing a mathematically robust healthy-state baseline designed for real-time condition assessment and online thermal diagnostics in mission-critical power conversion systems.
Insulated gate bipolar transistors, commonly abbreviated as IGBTs, represent core switching components deployed throughout modern high-capacity energy architectures, including flexible direct-current transmission networks, renewable utility-scale grid integration, and grid-forming power equipment. Prized for their high voltage-blocking capabilities, modest control power requirements, and relatively low conduction losses, these devices handle intense electrical stresses during continuous power delivery. Operational safety in commercial installations frequently relies on active voltage clamping schemes triggered by predefined on-state voltage-drop thresholds. However, conventional clamping protections are fundamentally designed to intervene during catastrophic short circuits or instantaneous current surges rather than gradual physical wear. Operational surveys indicate that approximately 55 percent of power module failures stem from temperature-related mechanisms. Junction temperature values and the amplitude of thermal cycling serve as primary health indicators for structural degradation, underscoring the urgent necessity of responsive, continuous junction temperature monitoring during ongoing power conversion.
Engineering methodologies for tracking IGBT junction temperatures generally encompass four traditional categories: direct physical contact sensors, optical non-contact instrumentation, thermal-network computational observers, and temperature-sensitive electrical parameter extraction. Among these alternatives, parameter extraction approaches estimate internal thermal states by establishing mathematical relationships with readily accessible electrical parameters, notably turn-off durations or the on-state voltage drop. The on-state collector-emitter voltage drop stands out as an exceptionally promising indicator due to its pronounced thermal sensitivity, low hardware acquisition cost, and minimal disruption to main power pathways. Nevertheless, mapping on-state voltage behavior across dynamic operational states has historically presented formidable theoretical and practical obstacles. Existing implementations typically depend either on dense look-up tables compiled from extensive laboratory characterization or on empirical polynomial equations fitted to narrow operational zones, leaving substantial performance gaps when operating conditions fluctuate unexpectedly.
A critical engineering deficiency in classical polynomial and data-driven representations is their fragile validity outside calibrated operating boundaries. Conventional polynomial formulations generally exhibit satisfactory estimation precision only within continuous power-conversion regimes operating below roughly twice the rated device current, typical of steady traction drives or photovoltaic inverters. However, modern grid-forming interfaces, high-voltage direct-current systems, and protective interruption equipment increasingly subject switching modules to short-term overloads and fault transients reaching four to five times nominal current ratings. In this high-current regime, complex nonlinearities between voltage drop and thermal states intensify significantly. When classical polynomial equations fitted around nominal ratings are extrapolated to extreme currents, their estimation errors escalate dramatically or break down into non-physical mathematical solutions. Furthermore, because traditional curve fits lack physical semiconductor foundations, any shift in gate voltage or operating baseline historically required laborious recalibration and parameter re-identification.
To overcome these structural limitations, the investigators constructed their closed-form model upon semiconductor physics, conceptualizing the IGBT as an equivalent circuit comprising a P-type–Intrinsic–N-type diode wired in series with a metal-oxide-semiconductor field-effect transistor operating in its linear conduction region. Under this physical decomposition, the total collector-emitter voltage equals the sum of the diode voltage drop and the channel voltage drop across the MOSFET. The internal MOSFET current-voltage relationship was formulated using an expression containing the gate-drive voltage, an equivalent channel threshold voltage parameter, a nonlinear gate overdrive exponent, and a scaling coefficient. By solving the resulting quadratic equation for the internal MOSFET voltage drop, the authors expressed channel conduction as an analytical function of the channel current, effectively bridging physical channel transport with externally measurable module parameters.
The mathematical formulation further accounted for the division of device current between electron channel pathways and bipolar conduction mechanisms. Under static operational states, the ratio of the internal MOSFET electron current to the overall collector-emitter current was modeled as proportional to the gate-drive voltage raised to the 1.5 power, scaled by a characteristic coupling factor. Integrating this current partitioning into the total voltage drop equation produced a unified relationship containing an overarching device correction constant to compensate for unmodeled structural parasitics. Because thermal fluctuations simultaneously modulate channel inversion, carrier transport, and bipolar region conductivity, the authors explicitly defined linear temperature dependencies for both the equivalent threshold voltage and the internal diode voltage drop. The threshold voltage incorporated a negative temperature coefficient reflecting enhanced thermal carrier excitation, whereas the diode drop incorporated both an ohmic slope factor and a positive temperature coefficient reflecting decreased carrier mobilities at elevated temperatures.
To ensure robust physical behavior and stable numerical convergence during model parameter identification, strict mathematical boundaries were assigned to all internal variables based on fundamental semiconductor properties. Because the electron and hole mobilities in silicon decrease substantially with increasing temperature near room levels—governed by power-law temperature dependencies with exponents of negative 2.42 and negative 2.20, respectively—the resistive voltage drop across the drift region increases under high current densities, mandating a strictly positive temperature coefficient for the equivalent diode voltage drop. Similarly, the ohmic proportionality parameter was constrained to positive values, while the threshold voltage temperature coefficient was held strictly negative. The gate overdrive exponent was constrained between negative one and zero to regulate nonlinear expansion, and the composite current-coupling term was restricted to positive values, ensuring the model faithfully emulated real physical conduction throughout the wide operating current regime.
Experimental validation of the theoretical baseline model required an electrical testing apparatus capable of generating extreme overload currents without inducing confounding self-heating within the test specimen. The researchers established a specialized resonant high-current circuit utilizing resonant energy exchange between a capacitor bank and an inductor. By precisely regulating the charging voltage across the capacitor bank, the test platform produced massive current pulses within a brief window of hundreds of microseconds, turning off the device immediately after peak current passage. Because the entire pulse occurred over an interval of approximately 150 microseconds, transient power dissipation within the semiconductor silicon die remained exceptionally low, producing an internal temperature rise negligible compared to ambient thermal chamber settings. Consequently, the temperature inside the environmental chamber, adjustable across a range of 25 to 105 degrees Celsius, served as the precise ground-truth reference for calibrating and evaluating junction temperatures.
The physical test platform evaluated a commercial press-pack trench-gate IGBT, specifically model ST750GXH24 manufactured by Toshiba, featuring a rated current of 750 amperes and the mechanical resilience to withstand short-duration overcurrent surges up to roughly 4.5 times its rated capacity. To capture clean waveforms under high common-mode voltages, the authors developed an integrated gate-drive and measurement circuit board partitioned into power supply, driver, and sensing modules. A specialized acquisition circuit utilized a complementary MOSFET switch and a blocking diode to electrically isolate sensitive operational amplifiers from lethal high-voltage transients during device turn-off, shunting excess current safely away from measurement paths. Fiber-optic transmitters triggered gate voltages ranging from 15 to 24 volts, allowing the team to characterize conduction behavior under standard 15-volt driving conditions as well as elevated gate biases employed to suppress conduction losses during extreme current overloads.
Model parameters were formally identified using experimental waveforms gathered under a standard gate-drive voltage of 15 volts across the full thermal range, employing nonlinear least-squares curve fitting to minimize residual errors. Once determined, all characteristic parameters remained completely fixed, and the model was evaluated at higher gate biases of 20 volts and 24 volts without re-optimizing parameters. The resulting coefficients of determination reached 0.99841 at 15 volts, 0.98474 at 20 volts, and 0.98222 at 24 volts, confirming that a single unified parameter set captured device physics across disparate control states. Root-mean-square voltage fitting errors were limited to 28.602 millivolts, 88.875 millivolts, and 91.641 millivolts for the respective gate voltages, with maximum absolute voltage residuals of 49.789, 189.490, and 243.780 millivolts. When converted to thermal predictions, the maximum absolute junction temperature estimation errors were just 1.50 degrees Celsius at 15 volts, 2.91 degrees Celsius at 20 volts, and 1.05 degrees Celsius at 24 volts.
To evaluate generalizability across diverse manufacturing architectures, the researchers applied their semi-empirical formulation to two additional high-power commercial devices: an FZ1200R33KF2 module operated at 15 volts and an MG600Q2YS60A module operated at 20 volts. Parameter fitting yielded coefficients of determination of 0.9804 and 0.9895, alongside voltage root-mean-square errors of 147.4 millivolts and 54.2 millivolts, respectively. The investigative team also benchmarked the baseline model against classical first-, second-, and third-order polynomial functions by training models solely on data below three times rated current and challenging them to extrapolate behavior up to 4.5 times rated current at 35, 55, and 75 degrees Celsius. While polynomial formulations showed acceptable interpolation within local training zones, their high-current predictions suffered severe condition-dependent errors or returned non-physical mathematical failures devoid of real-valued solutions, whereas the physics-based baseline sustained stable predictive performance throughout the extrapolation window.
To confirm engineering practicality in live environments, the authors constructed a standalone online monitoring instrument centered on an STM32 microcontroller minimum system board coupled to an organic light-emitting diode display. Conditioning circuits scaled and filtered analog collector-emitter voltage and current signals before feeding them into on-chip analog-to-digital converters sampling at one-microsecond intervals. Executing the preloaded closed-form baseline equations, the embedded microcontroller computed and refreshed junction temperatures with an end-to-end processing latency of approximately 60 microseconds, well within the 150-microsecond duration of typical high-current surge events. The authors emphasized that the present formulation serves strictly as a healthy-state baseline, omitting long-term parameter drifts caused by bond-wire fatigue, solder degradation, or chip aging, but providing the rigorous operational foundation necessary for future advanced degradation tracking in high-power semiconductor installations.
Subject of Research: Online junction temperature monitoring of high-power insulated gate bipolar transistors across wide current operating ranges
Article Title: A semi-empirical closed-form healthy-state baseline model for online junction temperature monitoring of high-power IGBTs across a wide current operating range
Article References: Wang, L., Li, X., Sun, H., & Liu, X. (2026). A semi-empirical closed-form healthy-state baseline model for online junction temperature monitoring of high-power IGBTs across a wide current operating range. Journal of Engineering and Applied Science, 73(1), Article 461. https://doi.org/10.1186/s44147-026-01325-8
Image Credits: AI Generated
DOI: 10.1186/s44147-026-01325-8
Keywords: insulated gate bipolar transistors, junction temperature monitoring, on-state voltage drop, power semiconductor devices, temperature-sensitive electrical parameters, semi-empirical modeling, press-pack IGBT, threshold voltage, overload currents, microcontroller implementation, STM32 microcontroller, thermal characterization, high-power electronics, equivalent circuit modeling, carrier mobility, closed-form baseline
Cite Scienmag News
Denise Maddox. (November 17, 2026). Physics-Based Baseline Model Tracks IGBT Junction Temperature Across Extreme Operating Currents. Scienmag. https://scienmag.com/physics-based-baseline-model-tracks-igbt-junction-temperature-across-extreme-operating-currents/
Denise Maddox. "Physics-Based Baseline Model Tracks IGBT Junction Temperature Across Extreme Operating Currents." Scienmag, 17 November 2026, https://scienmag.com/physics-based-baseline-model-tracks-igbt-junction-temperature-across-extreme-operating-currents/. Accessed 17 November 2026.
Denise Maddox. "Physics-Based Baseline Model Tracks IGBT Junction Temperature Across Extreme Operating Currents." Scienmag. November 17, 2026. https://scienmag.com/physics-based-baseline-model-tracks-igbt-junction-temperature-across-extreme-operating-currents/








