Online Junction Temperature Measurement of IGBT Power Modules with Chip-Integrated Sensor

Online Junction Temperature Measurement of IGBT Power Modules with Chip-Integrated Sensor 1st Matthias Meißner Chair of Power Electronics TU Dresden Dresden, Germany [email protected] 2nd Dirk Rudolph Chair of Power Electronics TU Dresden Dresden, Germany [email protected] 3rd Sebastian Rode Chair of Power Electronics TU Dresden Dresden, Germany [email protected] 4th Steffen Bernet Chair of Power Electronics TU Dresden Dresden, Germany stef[email protected] Abstract—Improving reliability and resource efficiency in power electronics increasingly relies on accurate assessment of ageing and precise estimation of the remaining service life. Accurate temperature measurement of power semiconductors during converter operation is essential for these assessments. In this paper, a method for precise monitoring of semiconductor temperature under real operating conditions is investigated, providing a foundation for improved lifetime estimation and reliability assessment. Index Terms—IGBT junction temperature, on-chip temperature sensing, Tj sensor, IGBT monitoring, condition monitoring I. INTRODUCTION The online monitoring of junction temperature is becoming increasingly important in power electronics due to rising power density and reliability requirements. To achieve long lifetime, high availability, and low costs, significant efforts are being made, for example by introducing design procedures based on the physics of failure, analyzing application profiles and using electro-thermal simulations to predict the service life of critical components [1], [2]. An established technique for determining the junction temperature is the VCE(ϑ)method. With this method, the forward voltage drop of the IGBT VCE in the onstate is measured at a low measuring current, typically around 1/1000 of the nominal collector current [3]. The measurement is then converted into the corresponding junction temperature ϑjusing a temperature-dependent calibration curve established beforehand. However, this approach cannot be realized during converter operation. Other temperature-sensitive electrical parameters (TSEPs), such as threshold voltage, gate plateau voltage, turnon and turn-off delays, and gate resistance, require complex evaluation circuits, as they must detect low signal-tonoise ratios in highly disturbed environments [4]. A further Funded by the European Union with the grant agreement number 101096387 and by the German Federal Ministry of Education and Research under grant number 16MEE0227. The views and opinions expressed herein are solely those of the author(s) and do not necessarily reflect those of the European Union or the German Federal Ministry of Education and Research. Neither the European Union nor the German Federal Ministry of Education and Research can be held responsible for them. (a) DC+ AC DC− ϑ G1 G2 E1 E2 Temp EE2 (b) Fig. 1. (a) Chip with integrated temperature sensor [6] and (b) schematic drawing of EconoDUAL™3 power module with integrated temperature sensor. limitation is that, due to their inherent operating principles, these methods can only capture discrete temperature measurements during specific switching events, such as when the IGBT is turned on or off, or, in the case of internal gate resistance measurements, only during the off-state of the IGBT in converter operation [5]. The characterization of onchip temperature measurements in other publications often describes the measurement behavior without active switching of the IGBT [6], [7]. One exception is [8], which also observes electrical coupling between the load circuit and the measuring circuit, but without analyzing the dependencies in more detail. In this paper, an EconoDUAL™3power module from Infineon Technologies with an integrated on-chip sensor is used for real-time junction temperature measurement. Its accuracy is evaluated through comparison with established methods, including the VCE(ϑ)method, thermocouples, and infrared thermal imaging. Results show that the sensor enables continuous and accurate temperature monitoring during operation, making it suitable for converter applications. However, it is shown that the measured temperature is influenced by the load current due to the sensor’s electrical integration into the emitter path. To address this limitation, a compensation method based on the current-dependent parasitic elements of the chip structure is introduced. The compensated measurements demonstrate improved reliability and consistency, highlighting the potential of on-chip sensing for advanced condition monitoring in power electronic systems. II. CHIP-INTEGRATED TEMPERATURE SENSOR The measurement principle of the chip-integrated temperature sensor is based on the temperature-dependent forward voltage of a pn junction. An example of a chip with an integrated temperature sensor is shown in Fig. 1a. The temperature is determined by applying a small constant current to the measuring diode shown in Fig. 1b and recording the resulting voltage drop. Even though Fig. 1a does not depict the actual chip used in this study, the temperature sensor in the employed chip is also integrated in the silicon and is located in the center. A. Experimental Setup and Data Preprocessing A self-developed control platform operating at 50 kHz was used to record the on-chip sensor and two Type-K thermocouples. The following description of averaging applies only to the measurements at room temperature and during the quasistationary cooling process described in Section II-B. For these sensors, each data point ¯ ϑrepresents the average of 1001 samples (sampling interval 20 ms) taken every 30 seconds. The Type-T thermocouple was recorded using a TC-08 data logger from Pico Technology at 10 Hz, with each temperature value representing the average of 100 samples (sampling interval 9 s) acquired every 30 seconds. Table I summarizes an overview of the measuring devices used. The measurement uncertainty of the Pico Technology data logger, used to record the Type-T thermocouple measurements, is given by ∆TPicoLog =±(0.2 % · |ϑ|+ 0.5◦C) (1) which defines the device’s temperature-dependent tolerance. The temperature tolerance of the Type-T Class 1 thermocouple is specified as ∆TTypeT ={︄±0.5◦C,−40 ◦C≤ϑ≤125 ◦C ±0.4 % ·ϑ, 125 ◦C< ϑ ≤350 ◦C(2) TABLE I MEASURING DEVICES AND DATA ACQUISITION. Measurement Device Acquisition on-chip temperature and Type-K thermocouples developed measurement PCBs and control platform control frequency 50 kHz Type-T thermocouple Pico Technology TC-08 Type-T Class 1 thermocouple 10 Hz infrared camera Optris PI640 125 Hz oscilloscope LeCroy HDO6054-MS 12 Bit;500 MHz; 2.5 GS/s load current during short pulses Rogowski coil PEM CWT6B mini 1.2 kA; 0.9 Hz –15 MHz on-chip voltage during short pulses passive probes LeCroy PP008 1:10; 500 MHz Fig. 2. Principal structure of a power semiconductor module, modified from [3] with added thermocouple positions for temperature measurement. depending on the measured temperature range. The total uncertainty of the Type-T measurement, combining the thermocouple and device tolerances, is calculated using Gaussian error propagation as ∆TTypeT,RMS =√︂∆T2 PicoLog + ∆T2 TypeT .(3) B. Static Temperature Measurement To determine the static measurement temperature offset of the sensors, initial measurements were carried out at room temperature. The temperature profiles of the on-chip sensor, one Type-T, and two Type-K thermocouples were recorded. As shown in Fig. 2, one of the Type-K and the Type-T sensor were positioned directly on top of the IGBT chip (without silicone gel) and the other Type-K was positioned on the side edge of the module’s base plate. The temper0 5 10 15 20 tin min ! 24 25 26 27 28 #in /C! 7 #OnChip 7 #TypeK Chip 7 #TypeK Case 7 #TypeT Chip (a) 0 5 10 15 20 tin min ! !1 0 1 2 3 Tin K ! 7 #OnChip !7 #TypeT Chip 7 #TypeK Chip !7 #TypeT Chip 7 #TypeK Case !7 #TypeT Chip (b) Fig. 3. Comparison of (a) the integrated temperature measurement with TypeK thermocouples at room temperature and (b) the difference between the values and the Type-T thermocouple on the chip. ature values derived from the sensor readings based on the manufacturer’s specifications are shown in Fig. 3a together with the thermocouple measurements. Fig. 3b shows the deviations between the Type-T thermocouple and both the onchip temperature measurement and the Type-K thermocouples at room temperature. The Type-K elements show temperature deviations of approximately −0.1 K and 1.2 K, both of which are within the specified tolerance of ±1.5 K for Class 1 TypeK thermocouples. The deviation of 1.4 K observed in the onchip measurement is considered acceptable, given that neither the sensing diode nor the developed measurement circuit were calibrated. Furthermore, the reference measurement using the Type-T thermocouple, as defined by (1) – (3), exhibits an uncertainty of ±0.74 K at 25 ◦C. To calibrate the VCE(ϑ)method, a quasi-stationary cooling process was recorded in which the system was thermally insulated and initially heated to a stable temperature. During the subsequent cooling process, the individual temperature measurement points can be regarded as stationary due to the low temperature gradient. At the same time, the voltage of the on-chip sensor was measured together with the data from the three thermocouples. Fig. 4a compares the integrated temperature measurement ¯ ϑOnChip with the Type-K thermocouples ¯ ϑTypeK and the Type-T thermocouple ¯ ϑTypeT during the cooling process over several hours. Fig. 4b shows the deviation between these measurements as a function of the temperature of the TypeT measurement. In addition, the uncertainty for the Type-T measurement ∆TTypeT,RMS according to (3) is shown. It 0 5 10 15 tinh! 40 60 80 100 120 140 #in /C! 7 #OnChip 7 #TypeK Chip 7 #TypeK Case 7 #TypeT Chip (a) 40 60 80 100 120 140 7 #TypeT Chip in /C! !1 0 1 2 3 Tin K ! 7 #OnChip !7 #TypeT Chip 7 #TypeK Chip !7 #TypeT Chip 7 #TypeK Case !7 #TypeT Chip "TTypeT RMS (b) Fig. 4. Comparison of (a) the integrated temperature measurement with TypeK thermocouples for the quasi-stationary cooling process in dependence of the time and (b) the difference between the values and the thermocouple on the chip in dependence of the temperature. illustrates that for temperatures between 35 ◦Cand 145 ◦C, the deviation of the on-chip temperature sensor is smaller than the specified uncertainty of the Type-T measurement. The deviations are greater at lower and higher temperatures and cannot be explained solely by a linearization error or an offset due to their course. Since the observed error remains small (less than 1.5 K) it is acceptable for further evaluation. Neither the offset nor the linearity of the sensor are corrected for further analysis in this study. Due to the minimal deviation between the measurements and the objective of validating the on-chip sensor under dynamic temperature profiles in the following sections, the VCE(ϑ)method was calibrated using the on-chip temperature as a reference. The resulting temperature ϑVce, estimated by this method, is approximated as a linear function of the measured collector-emitter voltage VCE: ϑVce =c1·VCE +c2≈ϑOnChip (4) where the coefficients c1and c2are derived from the static measurement data. To prevent mutual influence of the VCE(ϑ)method by chips connected in parallel within the module, the measurements were carried out on a single IGBT chip while all others were electrically isolated. Variations in thermal and electrical coupling can lead to differences in chip temperatures within the module. While no temperature deviation is expected under static calibration conditions, this effect becomes significant during dynamic temperature measurements. C. Dynamic Temperature Measurement After achieving a high level of agreement between sensor data in the static temperature measurements, further evaluations were conducted under more dynamic conditions. By applying a constant load current, the module was heated to a steady-state temperature due to the resulting power loss. After the current was switched off externally, the cooling behavior was recorded using an infrared thermal camera, the VCE(ϑ) method, and the on-chip sensor. For the infrared temperature measurement, the optris PI640 camera was used (accuracy of ±2 K or ±2 %, whichever is greater). Fig. 5 shows the temperature distribution on the chip area in steady-state condition, as measured by the infrared camera. The analysis yields an average temperature of 123.8◦Cand a maximum temperature of 141.2◦C. The maximum temperature in a defined smaller area was also analyzed. Fig. 6 shows that no bond wires cover the chip in this region, allowing the surface temperature to be recorded as the chip reference temperature, ϑIR reference. This is particularly important for dynamic cooling, as bond wires cool down more slowly than the chip itself due to their lower thermal dissipation capability. In the statically heated state, however, the validity of the reference temperature ϑIR reference = 136.8◦Cis questionable. Although the onchip sensor is located at the center of the chip, where slightly higher temperatures are expected, the observed difference of 16.1 K from the on-chip temperature ϑOnChip = 152.9◦Cis Fig. 5. Temperature measurement using an infrared camera and the temperature profile along the chip diagonal. Fig. 6. Temperature measurement approximately 40 ms after switching off using an infrared camera and the temperature profile along the chip diagonal. too large to be explained by this effect alone. Notably, the deviation between the maximum temperature in the infrared image ϑIR area max and the reference surface ϑIR reference is comparatively small, whereas the deviation from the on-chip temperature is significantly larger. This suggests that other influencing factors are present. A possible explanation is the dependency of the on-chip measurement on the load current, which is examined in more detail in the following sections. Fig. 7 presents the temperatures measured by the on-chip sensor and the thermal camera, along with the temperature calculated using the VCE(ϑ)method. The VCE(ϑ)method estimates a temperature of 137.4◦Cat the moment of current switch-off. However, the fact that this temperature almost exactly matches the values measured on the smaller chip reference surface by the thermal imaging camera is purely coincidental, as the VCE(ϑ)measurement represents a currentaveraged temperature over the entire chip area. Additionally, the VCE(ϑ)method applies a square-root-of-time approach to linearize the temperature curve within the time window of 400 µsto 40 ms before extrapolating it to the switch-off moment. The values displayed for t < 0 s are identical to those at t= 0 s, ensuring consistency in the presented data. In addition, the shading effect of the bond wires contribute to a deviation in the temperature measured by the thermal imaging camera. According to [9], it would have been expected that the mean temperature recorded by the infrared camera and the temperature obtained using the VCE(ϑ)method would closely align. However, it is unexpected that, before the load current is switched off, the on-chip temperature is significantly higher than the maximum temperature recorded by the thermal camera. At the moment of switch-off, it is noticeable that the on-chip temperature initially decreases more rapidly than the temperature determined by the VCE(ϑ)method. Additionally, it becomes evident that the thermal camera is not fast enough to accurately capture these dynamic temperature changes. In the steady-state cooled condition, however, the temperatures 0 5 10 15 tin s ! 50 100 150 #in /C! #OnChip #Vce #IR area max #IR reference #IR area mean (a) !0.15 !0.1 !0.05 40 60 80 100 120 140 160 #in /C! 0 0.05 0.1 tin s ! 9 9.05 9.1 #OnChip #Vce #IR area max #IR reference #IR area mean (b) Fig. 7. Comparison of temperatures measured by the on-chip sensor, infrared camera (average and maximum) with temperatures calculated using the VCE(ϑ)method for (a) the entire duration and (b) specific intervals. obtained from the different measurement methods show very good agreement. III. COMPENSATION OF ON-CHIP TEMPERATURE MEASUREMENT The temperature measurement obtained from the on-chip sensor appears to be inaccurate during the conduction phase of the load current. One possible explanation is the galvanic connection between the on-chip sensor and the emitter metallization on the chip, through which both the measurement and load currents flow. Fig. 8 presents a simplified schematic of an IGBT module with nparallel IGBT and diode chips, including the relevant stray inductances and resistances. The parasitic elements, such as those present at both the collector and emitter, have been consolidated for clarity. The schematic illustrates that the sensor voltage VOnChip is affected by the voltage drop across the parasitic resistance RσT2,1t and stray inductance LσT2,1t, caused by the emitter current flowing through these elements. Consequently, compensation for these effects is required to ensure accurate temperature measurement. In practical module operation, the emitter current of an individual chip cannot be directly measured. Instead, the load current ILoad at the AC-terminal, which is typically known, is used for compensation. Considering nparallel chips and a non-ideal current distribution factor kcamong the parallel con- RσT2,1g G2,1 E2,1 Temp EE2RσT2,1t LσT2,1t LσT2,1g LσD2,1 Dϑ DC– AC E2,2 G2,2 RσT2,2 LσT2,2 LσL1 RσL1 E2,n G2,n LσD2,n RσD2,n RσT2,n RσD2,2 LσT2,n LσLn RσLn LσL2 RσL2 ILoad LσD2,2 RσD2,1 VOnChip Fig. 8. Simplified schematic of an IGBT module with nparallel IGBT and diode chips, including the relevant (aggregated) stray inductances and resistances. duction paths, the voltage drop across the parasitic resistance RσT2,1t can be expressed as VRσT2,1t =RσT2,1t ·IeT2,1=RσT2,1t kc n·Iload (5) with can be simplified to VRσT2,1t =RComp ·ILoad (6) where the equivalent compensation resistance RComp combines the parasitic resistance RσT2,1t with the number of parallel conduction paths nand their non-ideal current sharing described by kc. The compensation for the stray inductance was carried out in the same manner, using an equivalent compensation inductance LComp according to VLσT2,1t =LσT2,1t ·dIeT2,1 dt =LComp ·dILoad dt. (7) A. Determination of Compensation Parameters To determine the compensation parameters RComp and LComp, the IGBT module was tested in a double-pulse setup with low load inductance (LLoad ≈3.5µH). This setup enabled the application of load currents in the range of the module’s nominal current using short conduction pulses, ensuring that chip heating remained negligible. Since the IGBT must be actively switched for the evaluation, a module with insulation gel was used. Due to the limited number of modules with chip-integrated temperature sensors, the parallel connection of the chips was not canceled. In this context, an ideal current distribution was assumed by setting the coupling factor to kc= 1, which introduces a simplification that may lead to minor deviations in the extracted parameters. To ensure the required high time resolution, the measurements were recorded using an oscilloscope. The waveforms of the load current and the on-chip temperature are shown in Fig. 9. In the uncompensated temperature response, it is evident that the temperature returns to its initial value immediately after the load current is switched off. This confirms the assumption that no significant heating of the chip occurs !50 0 50 100 150 tin 7s! 0 100 200 300 400 500 Iin A ! Ie (a) !50 0 50 100 150 tin 7s! 22 24 26 28 30 32 34 #in /C! #OnChip #OnChip Comp (b) Fig. 9. Oscilloscope measurement of (a) emitter current and (b) measured and compensated temperature curve. during the measurement. However, it can also be observed that significant noise in the measurement signal occurs during the switching transitions due to voltage and current changes in the load current path. By applying different DC-link voltages, resulting in varying current rise rates, the compensation parameters were determined and validated. The resulting compensated temperature ϑOnChip Comp is also shown in Fig. 9. The applied values for resistance and inductance compensation were RComp = 67 µΩ and LComp = 180 pH, respectively. B. Dynamic Temperature Measurement with Compensation Using the compensation parameters determined from short current pulses, it is possible to correct the temperature measurement during the conducting phase of the IGBT. Since high current change rates occur primarily during switching events, only the resistance-dependent component has been considered in the compensation. It should be noted that the calculated compensation resistance applies to the parallel connection of several chips. Since only one chip is used for the comparison with the VCE(ϑ)method, this was taken into account in the calculation. However, as the current distribution between the chips does not necessarily have to be symmetrical, there may be inaccuracies in the compensation. A comparison of the compensated on-chip temperature ϑOnChip Comp, the thermal camera measurements ϑIR, and the VCE(ϑ)method is presented in Fig. 10. 0 5 10 15 tin s ! 50 100 150 #in /C! #OnChip Comp #Vce #IR area max #IR reference #IR area mean (a) !0.15 !0.1 !0.05 40 60 80 100 120 140 160 #in /C! 0 0.05 0.1 tin s ! 9 9.05 9.1 #OnChip Comp #Vce #IR area max #IR reference #IR area mean (b) Fig. 10. Comparison of (a) temperatures measured by the current compensated on-chip sensor and infrared camera (average and maximum) with temperatures calculated using the VCE(ϑ)method for the entire duration and (b) specific intervals. It becomes evident that while the load current flows through the IGBT, the on-chip temperature reaches 148.5◦C, which is slightly higher than the maximum chip temperature of 141.0◦Crecorded by the thermal camera. This discrepancy of 7.5 K can be attributed to an imperfectly adjusted emissivity setting of the thermal camera, partial shading effects caused by the bond wires, and a suboptimally adjusted focus. During the quasi-stationary cooling phase (Section II-B), the temperature measured by the integrated sensor remained consistent with the values recorded by the Type-K and the Type-T thermocouples, supporting the validity of the on-chip measurements in the currentless state of the IGBT. It is therefore assumed that the thermal imaging camera underestimates the maximum temperature and that the temperature in the center of the chip actually corresponds to the on-chip sensor value. IV. CONCLUSION This paper demonstrates that the on-chip sensor can measure temperatures with deviations of less than 1.5 K in the absence of load current, even without calibration. However, during active switching of the IGBT, disturbances occur in the sensor signal due to transient voltage and current changes. By reducing the measurement bandwidth, it should be possible to filter out these disturbances without significantly impairing the dynamic response of the temperature measurement. In addition, it has been shown that despite load current compensation, the steady-state temperature in the conductive state deviates by 7.5 K from the maximum temperature recorded with the thermal imaging camera. Nevertheless, the on-chip sensor proves to be a very suitable method, as the preparation and measurement effort is significantly lower compared to other techniques, and the method can be applied during active converter operation. These findings highlight the potential of the on-chip sensor as a promising method for accurate and real-time temperature monitoring in power electronic applications. Future work should investigate how many on-chip sensors are necessary per power module to ensure reliable temperature monitoring. Measuring every individual chip is impractical due to cost and space constraints. A configuration with one sensor per switch (e.g., one for the low-side and one for the high-side IGBT), or a combination with a conventional NTC placed at the module edge, may be sufficient. Additionally, further studies should evaluate which aging mechanisms can be reliably detected using this method. While fatigue in the chip-solder interface is likely to manifest through thermal behavior, other failure modes, such as bond wire lift-off, may be more challenging to identify. REFERENCES [1] Huai Wang, Ke Ma, and Frede Blaabjerg. Design for reliability of power electronic systems. In IECON 2012 - 38th Annual Conference on IEEE Industrial Electronics Society, pages 33–44. IEEE, October 2012. [2] Sebastian Rode, Hulin Liang, and Steffen Bernet. A feedback-enhanced approach to modeling semiconductor ageing in electric vehicle inverters. The 26th European Conference on Power Electronics and Applications, 2025. [3] Infineon Application Note AN2015-10. Transient thermal measurements and thermal equivalent circuitmodels. V1.2 2020-04-14. [4] Sven Kalker, Lukas A. Ruppert, Christoph H. van der Broeck, Johannes Kuprat, Markus Andresen, Timothy A. Polom, Marco Liserre, and Rik W. De Doncker. Reviewing thermal-monitoring techniques for smart power modules. IEEE Journal of Emerging and Selected Topics in Power Electronics, 10(2):1326–1341, April 2022. [5] Michael Gleissner, Dominik Nehmer, and Mark-M. Bakran. Junction temperature measurement based on the internal gate resistance for a wide range of power semiconductors. IEEE Open Journal of Power Electronics, 4:293–305, 2023. [6] Eric R. Motto and John F. Donlon. Igbt module with user accessible onchip current and temperature sensors. In 2012 Twenty-Seventh Annual IEEE Applied Power Electronics Conference and Exposition (APEC). IEEE, February 2012. [7] Laurent Beaurenaut and Inpil Yoo. On-chip temperature measurement: a new approach for optimizing automotive inverter. In PCIM Europe 2016; International Exhibition and Conference for Power Electronics, Intelligent Motion, Renewable Energy and Energy Management, pages 1–8, 2016. [8] Badr-El-Boudour Bidouche, Yvan Avenas, Mouslim Essakili, and Laurent Dupont. Thermal characterization of an igbt power module with on-die temperature sensors. In 2017 IEEE Applied Power Electronics Conference and Exposition (APEC), pages 2317–2322. IEEE, March 2017. [9] Ralf Schmidt and Uwe Scheuermann. Using the chip as a temperature sensor — the influence of steep lateral temperature gradients on the vce(t)- measurement. In 2009 13th European Conference on Power Electronics and Applications, pages 1–9, 2009.