Thermal characterization of gallium nitride p-i-n diodes

Thermal characterization of gallium nitride p-i-n diodes J. Dallas, G. Pavlidis, B. Chatterjee, J. S. Lundh, M. Ji, J. Kim, T. Kao, T. Detchprohm, R. D. Dupuis, S. Shen, S. Graham, and S. Choi Citation: Appl. Phys. Lett. 112, 073503 (2018); doi: 10.1063/1.5006796 View online: https://doi.org/10.1063/1.5006796 View Table of Contents: http://aip.scitation.org/toc/apl/112/7 Published by the American Institute of Physics Articles you may be interested in Guest Editorial: The dawn of gallium oxide microelectronics Applied Physics Letters 112, 060401 (2018); 10.1063/1.5017845 Enhanced mobility in vertically scaled N-polar high-electron-mobility transistors using GaN/InGaN composite channels Applied Physics Letters 112, 073501 (2018); 10.1063/1.5010944 Tunnel-injected sub 290 nm ultra-violet light emitting diodes with 2.8% external quantum efficiency Applied Physics Letters 112, 071107 (2018); 10.1063/1.5017045 Optical AND operation in n-AlGaAs/GaAs heterojunction field effect transistor Applied Physics Letters 112, 072101 (2018); 10.1063/1.5010845 Temperature dependent electrical properties of pulse laser deposited Au/Ni/β-(AlGa)2O3 Schottky diode Applied Physics Letters 112, 072103 (2018); 10.1063/1.5019310 2 kV slanted tri-gate GaN-on-Si Schottky barrier diodes with ultra-low leakage current Applied Physics Letters 112, 052101 (2018); 10.1063/1.5012866 APPLIED PHYSICS LETTERS 112, 073503 (2018) Thermal characterization of gallium nitride p-i-n diodes J. Dallas,1 G. Pavlidis,2 B. Chatterjee,1 J. S. Lundh,1 M. Ji,3 J. Kim,4 T. Kao,3 T. Detchprohm,3 R. D. Dupuis,3 S. Shen,3 S. Graham,2 and S. Choi1,a) 1 Department of Mechanical and Nuclear Engineering, The Pennsylvania State University, University Park, Pennsylvania 16802, USA 2 George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332, USA 3 School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332, USA 4 LG Electronics, 38 Baumoe-Ro, Seocho-Gu, Seoul 06763, South Korea (Received 27 September 2017; accepted 2 February 2018; published online 14 February 2018) In this study, various thermal characterization techniques and multi-physics modeling were applied to understand the thermal characteristics of GaN vertical and quasi-vertical power diodes. Optical thermography techniques typically used for lateral GaN device temperature assessment including infrared thermography, thermoreflectance thermal imaging, and Raman thermometry were applied to GaN p-i-n diodes to determine if each technique is capable of providing insight into the thermal characteristics of vertical devices. Of these techniques, thermoreflectance thermal imaging and nanoparticle assisted Raman thermometry proved to yield accurate results and are the preferred methods of thermal characterization of vertical GaN diodes. Along with this, steady state and transient thermoreflectance measurements were performed on vertical and quasi-vertical GaN p-i-n diodes employing GaN and Sapphire substrates, respectively. Electro-thermal modeling was performed to validate measurement results and to demonstrate the effect of current crowding on the thermal response of quasi-vertical diodes. In terms of mitigating the self-heating effect, both the steady state and transient measurements demonstrated the superiority of the tested GaN-on-GaN vertical diode compared to the tested GaN-on-Sapphire quasi-vertical structure. Published by AIP Publishing. https://doi.org/10.1063/1.5006796 Gallium Nitride (GaN) has attracted attention for advancing legacy silicon (Si) based devices in high power and high frequency power electronic applications. Due to the superior material properties of GaN over Si and the resulting improvement in Baliga’s figure of merit,1 GaN devices show promise in developing power conversion systems with a reduced size and weight and improved power (SWaP) and efficiency.1–4 However, GaN application for next generation devices has previously been restricted by the lack of bulk GaN substrates forcing devices to be grown as lateral structures, including high electron mobility transistors (HEMTs). While lateral GaN structures have proven to be successful and have shown superior performance over Si devices, such devices are subject to performance restrictions preventing them from reaching the full potential of GaN. Such performance restrictions include limited breakdown voltage, dynamic on-resistance, current collapse, large residual stress, and overall larger chip size.5–8 However, more recent advancements in high quality bulk GaN substrates have enabled GaN vertical devices to be fabricated, such as diodes, which are not subject to the previously identified issues.9–11 Despite these structures possessing potential benefits, reduced device size and enhanced power handling capabilities result in intense self-heating which leads to several performance degradation attributes. For diodes, these include an exponential increase in leakage current at reverse biased conditions, increased specific on-resistance, increased a) Author to whom correspondence should be addressed: [email protected] 0003-6951/2018/112(7)/073503/5/$30.00 capacitance, breakdown voltage shift, and turn-on voltage shifts.5,10,12–14 Despite knowledge of this, limited studies have been performed to understand the thermal characteristics of GaN p-i-n diodes. Furthermore, thermal characterization techniques developed for lateral GaN structures are yet to be validated for use in GaN vertical devices. In this study, thermal analysis of vertical and quasivertical GaN p-i-n diodes was performed on the diodes that were fabricated as in Refs. 12 and 15 for the vertical and quasi-vertical diodes, respectively. Cross-sections of the tested devices are shown in Figs. 1(a) and 1(b). To determine if existing thermal characterization techniques can provide insight into thermal characteristics of vertical structures, steady state measurements with a base temperature of 45  C and forward bias power dissipating conditions of 200, 400, and 600 mW were performed on the quasi-vertical diode via FIG. 1. Simplified schematic of the tested (a) quasi-vertical and (b) vertical diodes. The diodes consisted of a 65 lm p-contact and an 85 lm GaN mesa diameter. The thicknesses of the p-GaN, i-GaN, n-GaN, and substrate were 280 nm, 6 lm, 1.1 lm, and 430 lm, respectively. 112, 073503-1 Published by AIP Publishing. 073503-2 Dallas et al. infrared (IR) thermography, thermoreflectance thermal imaging (TTI), and Raman thermometry. Infrared (IR) thermal microscopy was performed using a Quantum Focus Instruments InfraScope via a medium wavelength infrared (MWIR) InSb infrared camera with 512  512 pixels. A 15 MWIR objective was used to perform thermal imaging, resulting in a lateral resolution of 2.8 lm. Since measurements captured the temperature of the low emissivity p-contact, which can be subject to significant errors, the calibration procedure utilized an emissivity correction process similar to Ref. 16 based upon two base temperature sets of 35  C/55  C and 65  C/85  C. However, even with emissivity corrections, infrared measurements may result in errors due to poor signal-to-noise ratios resulting from the low emissivity of the p-contact and poor spatial resolution. Further information regarding inaccuracy of p-contact measurements can be found in the supplementary material. In addition, the infrared technique cannot accurately probe the GaN surface temperature, as GaN is transparent to infrared wavelengths and the measurement will be corrupted by signals originating from outside of the focal plane. As a result, thermoreflectance thermal imaging was utilized to validate infrared measurements on the p-contact since it is ideal for measuring highly reflective (and hence low emissivity) materials. Furthermore, thermoreflectance exhibits an enhanced lateral spatial resolution as compared to IR thermography. Thermoreflectance thermal imaging was performed on the p-contact using a Microsanj NT-210B system equipped with a 1626  1236 charge coupled device (CCD) camera. Experiments were carried out with a 20 objective (NA ¼ 0.35) and a variety of dispersed LED illumination sources including those with wavelengths of 470 nm and 530 nm that resulted in thermoreflectance coefficient (Cth) values17,18 with the largest magnitude. It should be noted that dispersed LEDs are used as opposed to laser based methods as the LEDs are of low enough power that heating effects from photon absorption by the metal can be neglected. The Cth values were determined by the process given in Ref. 19, which uses a temperature controlled stage to periodically elevate the temperature of the sample under study. Cth is determined over an iterative process that averages the difference in the reflectance signal captured at the elevated temperature (57  C) and the cooled temperature (25  C). Based upon this procedure, Cth values were determined to be 2.939  104 6 3.53  106 K1 and 1.788  104 6 1.62  106 K1 for green and blue light, respectively. The uncertainty associated with the Cth values mainly results from the rough and non-uniform surface of the p-contact, which leads to errors when averaging over the p-contact surface to determine a reliable value. Other sources of error include uncertainties associated with thermocouple reading and optical detection (for reflectivity calculation) during the Cth calibration process. The theoretical diffraction limited lateral spatial resolutions for both wavelengths were less than 0.75 lm, and the resolution of the detector was 0.18 lm/pixel. All diode timing/pulsing parameters were conservatively selected to ensure diodes to reach their steady state temperatures. The quasi-vertical diode was operated with square voltage pulses of 10% duty cycle and a 400 ls on-time, while the vertical diode was operated with 150 ls on-time for steady state Appl. Phys. Lett. 112, 073503 (2018) measurements. The duration of light illumination for the periodic measurement was 32 ls. While the thermoreflectance technique is ideal for capturing the temperatures of the reflective p-metal contact, it is rather incapable of performing reliable measurements on the GaN region.17 As such, Raman thermometry was performed as it has demonstrated success in measuring the temperatures of the channel region of GaN HEMTs. While existing Raman thermometry techniques, such as the multi-peak fit method,20 have demonstrated strength for temperature assessment of GaN HEMTs, they are limited in use on vertical devices due to obstructed optical access from the metallic p-contact and since measurements are carried out with a sub-bandgap laser to prevent induction of photocurrent. Measurements utilizing a sub-bandgap laser have been successfully applied to HEMTs since the GaN thickness is on the order of a few microns resulting in minimal depth averaging; however, vertical GaN structures consist of a relatively thick material stack composed almost entirely of GaN. This results in severe depth averaging and overlapping peaks originating from the differently doped GaN layers that are composed of the material stack, limiting the use of existing techniques based on linewidth and peak position shifts.21 Further information regarding depth resolution of confocal Raman thermometry can be found in the supplementary material. However, Raman thermometry is capable of assessing the surface temperature of vertical structures by applying an intermediate material to the surface such as nanoparticles, similar to measurements performed in Ref. 22. A survey of various nanoparticles was performed to find suitable candidates for nanoparticle assisted Raman thermometry including diamond (99% purity), aluminum nitride (99.5%), aluminum oxide (99.99%), boron nitride (BN) (99.8þ%), zinc oxide (99.5%), and titanium dioxide (99.9% and 99.98%). Each nanoparticle was selected because of its large bandgap, which minimizes heating from laser absorption and offers electrically insulating properties, since each material gives a distinct Raman peak. Of these, only anatase titanium dioxide (TiO2), diamond, and hexagonal boron nitride (BN) demonstrated sufficient signal-to-noise ratios in reasonable data accumulation times. The temperature FIG. 2. Raman temperature calibration of diamond, TiO2 (99.98%), and BN. TiO2 shows both higher sensitivity to temperature and lower uncertainty compared to BN and diamond. 073503-3 Dallas et al. Appl. Phys. Lett. 112, 073503 (2018) dependent Raman peak shifts of TiO2 (Eg mode: 143 cm1 at room temperature), diamond (F2g: 1328 cm1), and BN (E2g: 1367 cm1) are shown in Fig. 2. As can be seen, diamond and BN Raman peaks show similar sensitivity (slope) to temperature rise; however, BN shows a significantly less uncertainty. Additionally, TiO2 (99.98%) appears to be most sensitive to temperature changes as indicated by its large slope shown in Fig. 2 and Table I. Because of this, TiO2 was selected to be applied to the GaN p-i-n diodes to enable nanoparticle assisted Raman thermometry. The TiO2 nanoparticles were applied by dropping a solution of isopropanol and nanoparticles onto the diode. Isopropanol was then allowed to evaporate, leaving TiO2 on the diode surface. As this technique does not rely on plasmon resonance, such as in surface enhanced Raman spectroscopy (SERS), no additional fabrication steps in altering the surface roughness are needed. After applying TiO2 to the quasivertical diode, the nanoparticles were allowed to reach thermal equilibrium with the diodes and the Stokes peak shift of the Eg phonon mode23 was used to evaluate the temperature response of the nanoparticles. The application of nanoparticles creates several benefits as compared to traditional Raman techniques such as allowing surface temperature measurements of both the GaN and metal region, allowing lateral and vertical spatial resolutions to exceed that of its diffraction limit (in the case of probing isolated particles), and eliminating inclusion of stress effects in the Raman response to device temperature rise. Measurements were performed on the nanoparticles that were deposited on the p-contact. In order to reduce the overall Raman signal accumulation time and because of the lower material cost, agglomerated TiO2 particles with a purity of 99.9% were utilized instead of isolated TiO2 particles with a purity of 99.98%. Lateral x-y mapping measurements were performed using a Horiba LabRAM HR Evolution via a 2.5 lm diameter circular region with 0.25 lm steps. A sub-bandgap 532 nm laser was used to avoid particle-heating with a 10% neutral density filter and was transferred through a 50 objective, resulting in 8 mW laser power at the sample surface. A 15% electron multiplying charge coupled device (EMCCD) gain was also applied to amplify the signal of the TiO2 nanoparticles to reduce the signal acquisition time while maintaining the signal-to-noise ratio at a reasonable level. It was found that slopes correlating the peak position shift with the temperature rise from x-y mapping of agglomerated particles with a purity of 99.9% and individual particles with a purity of 99.98% show excellent agreement with each other. A more detailed discussion of nanoparticle purity and temperature gradient formation is given in the supplementary material. Thermoreflectance measurements were carried out on the GaN-on-GaN vertical diode, and the results were validated through simulation. A coupled electro-thermal modeling scheme24 using Sentaurus TCAD and COMSOL Multiphysics was utilized to estimate the heat generation and temperature distributions to gain further insight into the self-heating effect associated with each diode. For each mesh point, this model self-consistently solves the Poisson, current continuity, electrohydrodynamic, and heat conduction equations to derive the electrostatic potential, electron/hole concentration/energy/temperature distributions, heat generation, and lattice temperature rise. The resulting set of differential equations are discretized and coupled nonlinearly, and the solutions are obtained by a nonlinear iteration method. Figure 3(a) shows the radial temperature distribution of both the quasi-vertical and vertical diodes obtained through thermoreflectance thermal imaging. As can be seen, the vertical diode operates at a significantly lower temperature at a power dissipation of 600 mW due to the higher thermal conductivity of GaN compared to Sapphire. The p-contact lateral temperature distribution of the quasivertical diode is relatively uniform, while the vertical diode has a temperature more concentrated towards the center of the diode, decreasing towards the peripheral of the p-contact. Figure 3(b) compares the total heat generation cross-section TABLE I. Slope and uncertainty of the linear fit obtained for TiO2 (99.98%), BN, and diamond as obtained from Raman temperature calibrations. The typical acquisition time to obtain a signal-to-noise ratio greater than 20:1 at 35  C for agglomerated particles is also shown. Material Slope (cm1/ C) Error (cm1/ C) Acquisition time for a S/N of 20:1 TiO2 BN Diamond 0.02747 0.01932 0.01669 6.62E-04 7.59E-04 0.00281 1.9 15 >150a a Laser power was 4 mW, while others were performed at 8 mW. FIG. 3. (a) Radial temperature distribution at 600 mW obtained from thermoreflectance thermal imaging. Excursions in the measured temperatures originate from p-metal surface defects such as black spots that can be seen in the inset. (b) Heat generation profile obtained through electro-thermal modeling of the quasi-vertical structure and the vertical device. 073503-4 Dallas et al. obtained through electro-thermal modeling of the two diode structures. As can be seen, the quasi-vertical diode has a large heat generation concentrated near the periphery of the pcontact, extending vertically throughout the mesa.14 This is attributed to the phenomenon of current crowding where the electrons get crowded near the n-electrode and result in a spike in electron density near the p-electrode edge. A recent study14 on quasi-vertical and vertical diodes shows that this problem is significantly alleviated in the vertical design. As the vertical diode exhibits a more uniform current distribution and heat generation throughout the entire diode, the problem of current crowding is non-existent there. This suggests that the vertical diode structure is preferred over the quasi-vertical diode in terms of device reliability since the possibility of forming local hotspots under forward biased conditions is mitigated. Figure 4 shows the combined results of IR, thermoreflectance, Raman, and thermal simulation for the quasivertical diode. Overall, all techniques show reasonable agreement with thermal modeling. However, it is seen that at power levels of 200 and 400 mW, IR measurements underpredicted the temperatures compared to those obtained from thermal modeling. This is consistent with the fact that the p-contact metal has low emissivity. Essentially, this low emissivity causes low power (i.e., low temperature) measurements to be subject to an error resulting from low MWIR (2–4.5 lm) photon emission (or thermal radiation) of the metal surface. However, at 600 mW, the diode is of sufficiently high temperature that enough photons are being emitted to lessen the severity of errors to an extent that good agreement was achieved with the other techniques. Contrarily, mean values from Raman and thermoreflectance results show excellent agreement with each other and the model for all tested power levels. The large errors associated with the Raman measurements are due to error propagation from uncertainties associated with the x-y map based calibration method previously discussed. The agreement between these two techniques provides support for the accuracy of FIG. 4. Comparison of experimental and simulated temperatures of the p-contact of the quasi-vertical diode at forward power dissipation levels of 200, 400, and 600 mW with a base temperature of 45  C. For better readability, the IR thermography and thermoreflectance thermal imaging (TTI) data points were offset along the x-axis. As the uncertainty associated with thermoreflectance measurements is small (<0.66  C), the error bars are not visible in the graph. Appl. Phys. Lett. 112, 073503 (2018) the thermoreflectance results. Along with this, thermoreflectance demonstrated self-consistency among the tested illumination wavelengths—470 nm and 530 nm and white light (the results are not displayed in Fig. 4). While steady state measurements can provide insight into the heat generation profile of the diodes, these devices are normally operated as rectifiers, switching between forward and reverse biased states, and thus, understanding the transient thermal behavior of the diodes is important. Because of this, thermoreflectance transient measurements were performed with a temporal resolution of 3 ls. The transient measurements were performed at room temperature, while the diodes were switched from a zero-bias state to a forward bias condition dissipating 600 mW for both the vertical and quasi-vertical structures. Figure 5 shows the results of transient thermoreflectance measurements. As can be seen, the GaN-on-GaN vertical diode has a favorable transient response, as it operates at a lower peak temperature (due to the higher thermal conductivity of GaN as compared to Sapphire) and has a much faster cooling response. The thermal time constants were obtained through fitting the exponential decay and were determined to be 8.3 ls and 22.8 ls for the vertical and quasi-vertical diodes, respectively. This 65% improvement is thought to be mainly due to the removal of the thermal boundary resistance at the GaN/sapphire interface and the lower specific heat and thermal diffusivity of the GaN substrate compared to sapphire. As a result of this, the tested vertical diode is preferred over the tested quasi-vertical diode for high frequency operation, as it will have less residual thermal energy from the previous cycle remaining. Contrarily, the tested quasi-vertical diode will not have sufficient time when operating in high frequency ranges to thermally stabilize. This will result in an accession of the operating temperature of the diode, which may result in performance degradation attributes and impact the device reliability. Through the use of infrared thermal microscopy, thermoreflectance thermal imaging, Raman thermometry, and electro-thermal simulation, an accurate measurement scheme for GaN p-i-n diodes was established. It was demonstrated FIG. 5. Transient thermoreflectance measurement results for the vertical (red) and quasi-vertical (black) diodes at room temperature and 600 mW. 073503-5 Dallas et al. that infrared measurements were subjected to inaccuracy when assessing the temperature of the low emissivity p-contact at low power conditions. Because of this, the application of infrared thermography for temperature measurements of vertical devices should be limited for high power dissipating conditions where the metal (with emissivity of 0.15) temperature exceeds 100  C. If the goal of temperature mapping is to locate hotspots or failure points, IR thermography provides a convenient means to conduct such evaluation. Thermoreflectance and nanoparticle assisted Raman thermometry proved to be viable options for the temperature assessment of GaN p-i-n diodes. This suggests that the proposed measurement techniques can be applied to vertical WBG structures to further understand heat generation in such devices so that thermally aware designs can be exploited to allow the GaN vertical device technology to reach the full potential of the material. Along with this, steady state and transient measurements demonstrated the superior thermal characteristics of the tested vertical GaNon-GaN structure as compared to its GaN-on-Sapphire quasivertical counterpart. See supplementary material for considerations regarding TiO2 nanoparticle purity, accuracy of infrared thermography results for p-contacts, depth resolution of confocal Raman thermometry, and temperature gradient formation in nanoparticles during testing. Funding for efforts by the Pennsylvania State University was provided by the AFOSR Young Investigator Program (Grant No. FA9550-17-1-0141, Program Officer: Dr. Michael Kendra, also monitored by Dr. Kenneth Goretta). 1 B. J. Baliga, IEEE Electron Device Lett. 10, 455 (1989). J. Millan, P. Godignon, X. Perpina, A. Perez-Tomas, and J. Rebollo, IEEE Trans. Power Electron. 29, 2155 (2014). 3 T. P. Chow and R. Tyagi, IEEE Trans. Electron Devices 41, 1481 (1994). 4 H. A. Mantooth, M. D. Glover, and P. Shepherd, IEEE J. Emerging Sel. Top. Power Electron. 2, 374 (2014). 2 Appl. Phys. Lett. 112, 073503 (2018) 5 D. Disney, H. Nie, A. Edwards, D. Bour, H. Shah, and I. C. Kizilyalli, in Proceedings of the International. Symposium on Power Semiconductor Devices ICs (IEEE, 2013), pp. 59–62. 6 S. Karmalkar and U. K. Mishra, IEEE Trans. Electron Devices 48, 1515 (2001). 7 S. Choi, E. Heller, D. Dorsey, R. Vetury, and S. Graham, J. Appl. Phys. 113, 93510 (2013). 8 S. Choi, E. Heller, D. Dorsey, R. Vetury, and S. Graham, J. Appl. Phys. 114, 164501 (2013). 9 T. G. Zhu, D. J. H. Lambert, B. S. Shelton, M. M. Wong, U. Chowdhury, H. K. Kwon, and R. D. Dupuis, Electron. Lett. 36, 1971 (2000). 10 I. C. Kizilyalli, A. P. Edwards, H. Nie, D. Disney, and D. Bour, IEEE Trans. Electron Devices 60, 3067 (2013). 11 M. Trivedi and K. Shenai, J. Appl. Phys. 85, 6889 (1999). 12 T.-T. Kao, J. Kim, Y.-C. Lee, A. F. M. S. Haq, M.-H. Ji, T. Detchprohm, R. D. Dupuis, and S.-C. Shen, IEEE Trans. Electron Devices 62, 2679 (2015). 13 R. Raghunathan, D. Alok, and B. J. Baliga, IEEE Electron Device Lett. 16, 226 (1995). 14 X. Zhang, X. Zou, X. Lu, C. W. Tang, and K. M. Lau, IEEE Trans. Electron Devices 64, 809 (2017). 15 T.-T. Kao, Y.-C. Lee, M.-H. Ji, T. Detchprohm, R. D. Dupoi, and S. Shen, in Proceedings of the International Conference on Compound Semiconductor Manufacturing Technology (CS MANTECH) (2014), pp. 157–160. 16 C. H. Oxley, R. H. Hopper, G. Hill, and G. A. Evans, Solid State Electron. 54, 63 (2010). 17 K. Maize, E. Heller, D. Dorsey, and A. Shakouri, in 2012 28th Annual IEEE Semiconductor Thermal Measurement and Management Symposium. (IEEE, 2012), pp. 173–181. 18 K. Maize, G. Pavlidis, E. Heller, L. Yates, D. Kendig, S. Graham, and A. Shakouri, in 2014 IEEE Compound Semiconductor Integrated Circuit Symposium (IEEE, 2014), pp. 1–8. 19 A. Shakouri, A. Ziabari, D. Kendig, J.-H. Bahk, Y. Xuan, P. D. Ye, K. Yazawa, and A. Shakouri, in 2016 32nd Thermal Measurement Modelling and Management Symposium (IEEE, 2016), pp. 128–132. 20 S. Choi, E. R. Heller, D. Dorsey, R. Vetury, and S. Graham, IEEE Trans. Electron Devices 60, 1898 (2013). 21 T. Beechem, S. Graham, S. P. Kearney, L. M. Phinney, and J. R. Serrano, Rev. Sci. Instrum. 78, 61301 (2007). 22 J. Anaya, S. Rossi, M. Alomari, E. Kohn, L. T oth, B. Pecz, and M. Kuball, Appl. Phys. Lett. 106, 223101 (2015). 23  M. J. Sćepanović, M. Grujić-Brojčin, Z. D. Dohčević-Mitrović, and Z. V. Popović, Sci. Sintering 41, 67 (2009). 24 B. Chatterjee, J. S. Lundh, J. Dallas, H. Kim, and S. Choi, in IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (Orlando, 2017).