IEEE Xplore URL:

This paper has been accepted for publication by 2017 IEEE Applied Power Electronics Conference and Exposition, IEEE APEC. Personal use is permitted, but republication/redistribution requires IEEE permission. DOI: 10.1109/APEC.2017.7930963 Citation: S. Park and J. Rivas-Davila, "Power loss of GaN transistor reverse diodes in a high frequency high voltage resonant rectifier," 2017 IEEE Applied Power Electronics Conference and Exposition (APEC), Tampa, FL, USA, 2017, pp. 1942-1945. IEEE Xplore URL: http://ieeexplore.ieee.org/document/7930963/

Power Loss of Gate-Source-shorted GaN Transistors in Tens of Megahertz and Hundreds of Volts Class-DE Resonant Rectifiers Sanghyeon Park and Juan Rivas-Davila Electrical Engineering Stanford University Stanford, CA 94305 Email: spark15@stanford.edu Abstract This paper presents power loss measurements of gate-source shorted GaN transistors in place of diodes for high frequency high voltage rectification. To evaluate the performance, we use gate-source-shorted GaN transistors as passive rectifying devices in a class-de resonant rectifier and operate the circuit at 10s of megahertz switching frequencies and 100s of volts output voltages. Thermometric calibration method identifies power loss in all GaN transistors that increases with switching frequency and the rectifier output voltage. Furthermore, comparisons between experiments and simulations suggest that the diode power loss is neither unintended hard switching loss nor conduction loss from the forward voltage drop and on-resistance of the device. The correlation between the amount of power dissipation and GaN transistor output capacitance suggests that the observed power loss is related to output capacitance. I. INTRODUCTION High-frequency high-voltage dc-dc conversion makes possible small and lightweight power supplies and thus many exciting new applications [1] [4]. Efficient rectifying devices are critical in building high efficiency dc-dc conversion systems. This paper evaluates the performance of gate-sourceshorted Gallium Nitride (GaN) transistors as passive rectifying devices. Silicon (Si) and Silicon Carbide (SiC) Schottky diodes, both widely used in power systems, have their shortcomings when it comes to high-frequency high-voltage converter design. Si Schottky diodes have limited reverse voltage capability. SiC diodes, despite their high blocking voltage and low reverse leakage current, does not prove ideal due to a power loss that rapidly increases with the voltage and switching frequency [5]. Commercially available GaN transistors behaves like a diode when its gate node is shorted to the source node. Source and drain nodes correspond to diodes anode and cathode, respectively, in terms of their functionality. Manufacturers claim that this diode-like behavior comes with zero reverse recovery loss [6], [7]. Even though those gate-source-shorted GaN transistors forward voltage drop is several times larger than that of Si or SiC diodes, the corresponding power loss would be insignificant for applications that demand output voltage levels of hundreds of volts or higher [1] [4]. The GaN transistor with such configuration would ideally operate as an efficient passive rectifying device if we minimize switching losses by adopting a resonant rectifier topology.. This paper investigates GaN transistors performance in high frequency high voltage rectifiers, and compare the results with SiC diode counterparts. We use gate-source-shorted GaN transistors as a passive rectifying switch in class-de resonant rectifiers [8] (the referenced paper calls them class- D resonant rectifiers). We measure diode power losses by thermometric calibration and map device temperatures and power losses in a one-to-one fashion. The experiment identifies a significant loss from GaN transistors under tens of megahertz and hundreds of volts operation. We observe power losses increasing with the voltage and switching frequency, which we claim is attributable to neither conduction loss nor switching loss, calling for further investigation. The rest of the paper is organized as follows. In Section II we explain how the class-de resonant rectifier operates and describe the voltage and current waveforms that are applied to rectifying devices. In Section III we discuss the thermometric calibration method that we use to estimate the power losses in a device. In Section IV we show the measured GaN transistor losses under various voltage, frequency, temperature and current conditions. We also compare GaN transistors with SiC diodes in regard to their loss levels under the same conditions. In Section V we discuss possible mechanisms for the observed GaN losses. II. CLASS-DE RESONANT RECTIFIER We use class-de half bridge resonant rectifiers [8] illustrated in Fig. 1a to evaluate power losses in GaN transistors. The circuit includes an LC parallel tank in which the resonant capacitance is the sum of C res and diode junction capacitances C j,dgnd and C j,dout. The ac sinusoidal current input i s (t) resonates the LC parallel tank to generate a large voltage swing across diode D gnd. This voltage swing is clipped either to ground by diode D gnd or to the output voltage V out by diode D out. The subsequent current pulse flows from ground through D gnd and through D out to reach the output node, resulting in the dc output current I out.

i s (t) C block L res C res C out R load C j,dgnd D gnd D out C j,dout AC input LC tank Load + V out summarizes voltage and current ratings of the selected parts. III. THERMOMETRIC CALIBRATION v D + i D v D + GaNv D,pp = V out SiC i D impedance matching L Cblock ac input Lres v D Dout i D i D,avg=I out 1/f s t t dc output (c) Rload Fig. 1: Class-DE resonant rectifier we use to evaluate power loss in GaN transistors. Schematic of the rectifier circuit. The ac current input resonates LC parallel tank until the amplified voltage is clipped either to ground by diode D gnd or to V out by diode D out. The voltage across and current through diodes in the rectifier. We use SiC diodes or gate-source-shorted GaN transistors as a passive rectifying device. Provided that there is no reverse recovery transient, the resonant operation is ideally free of switching loss. (c) One of our rectifier implementations, a 500 V 27.12 MHz rectifier with a pair of gate-source-shorted GaN transistors used as passive rectifying diodes. Fig. 1b depicts the voltage across and current through diodes when the rectifier is operating at switching frequency f s. We use gate-source-shorted GaN transistors as rectifying devices and model them as a diode. v D,pp, the peak-to-peak voltage swing across the diode, is the same as the rectifier output voltage V out. The average diode current i D,avg is equal to the rectifier output current I out. As illustrated in the diode voltage v D and current i D waveforms, the circuit achieves zero voltage turn-on and zero current turn-off. Provided that we use diodes that do not exhibit any reverse recovery, the class-de resonant rectifier should be free of switching loss. Fig. 1c shows one of our rectifier implementations using GaN transistors. The circuit has output voltage 500 V and switching frequency 27.12 MHz with a pair of gate-source-shorted GaN transistors being used as passive rectifying diodes. The impedance matching inductor (not shown in Fig. 1a) matches the input impedance of the rectifier with the output impedance of the RF power amplifier that provides the ac input signal. We evaluate power losses of the following three GaN transistors: GaN Systems egan transistor; Navitas N V 6110 egan transistor; and Transphorm T P H3002LS cascoded dgan transistor. We also evaluate the following SiC Schottky diodes in the identical setup for comparison: STMicroelectronics ST P SC406B; Cree C3D04060E, CSD04060E, and C3D1P 7060Q. For the fairest possible comparison we selected parts of which voltage and current ratings are as similar to each other as possible. Appendix A Cres Dgnd Cout Fig. 2: Diode power loss estimation by thermometric calibration A constant current is injected to the diode while the device voltage and temperature are measured. Repeating this measurement multiple times yields the temperature vs. power loss calibration curve. Once the diode temperature is measured during the rectifier operation, the diode power loss is read from the calibration plot. We measure the power loss of rectifying devices in class- DE resonant rectifiers by thermometric calibration illustrated in Fig. 2. This method estimates the power dissipation of the device by the temperature it reaches in thermal equilibrium. Calibration curves correlate the temperature and power dissipation to enable loss estimation. We obtain the temperature versus power calibration curve as follows. First, we attach a populated rectifier circuit board onto a large metal block that maintains a constant temperature throughout the experiment. Second, we use a constant current source to inject a dc current into the rectifying device and record the temperature of the hottest spot on the device surface. We repeat this step for multiple power levels until the number of data points become large enough to draw a linear calibration curve. Finally, we operate the rectifier and measure the device temperature when the circuit reaches thermal equilibrium. We estimate the device power loss by using the calibration curve obtained above. The direct use of voltage and current probes, the most straightforward measurement method, is inappropriate in this study. Since this study focuses on rectifiers with tens of MHz switching frequency, capacitance and inductance in the circuit are only several tens of pf and few hundreds of nh. Voltage and current probes would add capacitance and inductance large enough to disrupt the circuit behavior significantly. IV. POWER LOSS MEASUREMENTS In this section, we show the power loss observed in GaN transistors and identify factors affecting loss levels. We also

3W 1W C3D04060E CSD04060E C3D1P7060Q STPSC406B 10.2 W 0V 200V 400V 600V Device Peak-to-peak Voltage 1 C3D04060E CSD04060E C3D1P7060Q STPSC406B 0MHz 20MHz 40MHz Switching Frequency 19.4 W 1 13W 1 11W 1 3W 1W 20 30 40 50 60 70 Case Temperature Fig. 3: Power losses in rectifying devices under various voltage, frequency and temperature conditions. Blue and red markers indicate GaN transistor losses and SiC diode losses, respectively. Power losses at 170 V, 350 V and 500 V of peak-to-peak voltage across the device. The average diode current (equivalent to the rectifier output current) is 50 ma. Power losses at 13.56 MHz, 27.12 MHz and 40.68 MHz of switching frequency. The peak-to-peak diode voltage is 500 V. (c) Power losses at case temperatures from 30 to 60 C. We change the device case temperature by manipulating the burst duty cycle. The peak-to-peak diode voltage is 500 V and average diode current is 50 ma. Note the broken vertical axis. (c) compare the losses of GaN with those of SiC Schottky diodes under identical conditions for reference purposes. Except for the power loss vs. temperature experiment, we maintain the temperature of the hottest spot of the device surface to be within 27-30 C by adjusting the rectifier burst duty cycle. A. Power Loss vs. Voltage, Frequency and Temperature Fig. 3 shows power losses in rectifying devices at various voltages, frequencies and temperatures. Plots show GaN transistor losses as blue markers and SiC diode losses as red markers. Fig. 3a illustrates power losses measured at 170 V, 350 V and 500 V of peak-to-peak voltage swing across GaN transistors and SiC diodes. Here all the data are at 27.12 MHz switching frequency. Fig. 3b displays power losses at 13.56 MHz, 27.12 MHz and 40.68 MHz of switching frequency, when the peak-to-peak voltage swing is 500 V. Fig. 3c shows the power losses in all three GaN transistors increase with the case temperature. We change the device case temperature from 30 to 60 C by manipulating the burst duty cycle of the rectifier. The peak-to-peak diode voltage is 500 V and average diode current is 50 ma. Note the broken vertical axis between 4 W and 10 W. Gate-source-shorted GaN transistors turn out to be inappropriate as passive rectifying devices in 100s of volts and 10s of megahertz operations. GaN power losses are at best comparable to SiC diode counterparts, and increase rapidly with voltage and frequency. Power losses per GaN device are significant in all cases in Fig. 3, ranging from 5.6 % (in N V 6110 at 170 V and 27.12 MHz) to 77.6 % (in T P H3002LS at 500 V and 40.68 MHz) of the rectifier output power. B. Power Loss vs. Fig. 4 shows power losses measured when the dc current through and N V 6110 GaN transistors varies from 50 ma to 1000 ma. Plots from Fig. 4a to Fig. 4d represent the power losses against the dc current through devices. In all voltage and frequency conditions, the power loss increases with the dc current through the device, although not as rapidly as with voltage and frequency. Plots from Fig. 4e to Fig. 4h rearrange the data to compare them with simulated GaN losses. The thick gray dashed line across each plot indicates the place where measured and simulated losses are equal. We obtain simulated losses by measuring the IV curve of gate-source-shorted GaN transistors and modelling them as ideal diodes in series with a constant voltage source and a linear resistor. In order to prevent the device temperature from affecting the IV curve measurement, we adjust the duty cycle of the dc current injection so that the temperature falls into the same range as in the device loss measurement experiment. In all four plots, the array of data points from the bottom left to the top right correspond to the increasing current levels from 50 ma to 1000 ma. Plots show that the additional power loss on top of the simulated loss (the gap between blue markers and the thick gray line) is a very weak function of the device current and constitutes a significant portion of total loss only at low current levels. At high dc current levels, most of the power loss is explained as a conduction loss and well-predicted by the aforementioned simple diode model. V. DISCUSSION ON LOSS MECHANISM We cannot explain the GaN power loss as the conduction loss predictable from the device IV curve. Looking at plots from Fig. 4e to Fig. 4h, following characteristics are apparent in all frequency and voltage conditions: the difference between simulated and experimental data sets remains nearly constant throughout the whole current range; the difference between those two increases with switching frequency and the

Voltage Across Device 1 1 1 1 1 2 1 1 at 27.12 MHz and various voltages at 500 V and various frequencies (c) NV 6110 at 27.12 MHz and various voltages (d) NV 6110 at 500 V and various frequencies 1 1 Simulated (e) compared with simulation at 27.12 MHz and various voltages 1 1 1 1 Simulated (f) compared with simulation at 500 V and various frequencies 1 1 1 1 Simulated (g) NV 6110 compared with simulation at 27.12 MHz and various voltages 2 1 1 1 1 2 Simulated (h) NV 6110 compared with simulation at 500 V and various frequencies Fig. 4: We vary the dc current through and NV 6110 GaN transistors from 50 ma to 1000 ma and measure the power loss at various voltages and frequencies. The switching frequency is 27.12 MHz for, (c), (e) and (g). The peak-to-peak device voltage is 500 V for, (d), (f) and (h). Power losses of at 27.12 MHz frequency and various peak-to-peak device voltages. Power losses of at 500 V peak-to-peak device voltage and various switching frequencies. (c-d) Power losses of NV 6110 under the same condition as in and. (e-f) Power losses of compared with the losses in simulation. (g-h) Power losses of NV 6110 compared with the losses in simulation. peak-to-peak voltage across the device. These characteristics indicate the existence of a power loss mechanism that is not a conduction loss. We cannot model the loss mechanism as a constant voltage drop and a linear resistance, or more complex models that reflect IV characteristics of GaN at low frequencies. Hard switching loss is also unlikely to be the mechanism of the observed GaN transistor power loss. Fig. 5 shows very good match between simulated (dashed lines) and experimental (colored solid lines) waveforms of the voltage across the device during rectifier operations at 27.12 MHz and three different output voltages. The good match between two voltage waveforms confirms that the rectifier circuit indeed operates as intended; the resonant operation recycles energy stored in junction capacitors, and zero-voltage switching occurs. Therefore, it is doubtful that the reason for the discrepancy between experimental and simulated data lies in hard switching. We suspect the power loss in GaN transistors is caused by the lossy output capacitance of the device. Fig. 6a shows that the output capacitance (C oss ) profile of [9] is almost 2.3 times that of N V 6110. Fig. 6b and Fig. 6c compares the power loss of at various voltages and frequencies which is also around 2.3 times that of NV 6110 in every case. This comparison suggests that the observed losses in GaN transistors might be proportionally related to C oss, or 600V 400V 300V 200V 100V 0V simul. -100V 0ns 37ns 74ns 111ns Time Fig. 5: Voltage waveforms across GS66052B GaN transsitor at 50 ma dc current, 27.12 MHz switching frequency and peak-to-peak voltages of 170 V, 350 V and 500 V. Good match between the experimental and simulated waveforms indicate that the hard switching loss is not the culprit for the GaN power loss. more generally, the semiconductor die area. VI. CONCLUSION This paper presented power losses of gate-source-shorted GaN transistors in place of rectifying diodes in a highfrequency high-voltage class-de resonant rectifier. All three

Max. Continuous Current Output Capacitance 3W 1W 125pF 100pF 75pF 50pF 25pF, x2.3 scaled 0pF 0V 100V 200V 300V 400V Drain-to-source Voltage, x2.3 scaled Device Peak-to-peak Voltage, x2.3 scaled Switching Frequency Fig. 6: Comparison of output capacitance C oss and power losses in and NV 6110 GaN transistors. C oss vs. source-to-drain voltage. The output capacitance profile of NV 6110, when scaled up by a factor of 2.3, almost coincides with that of. (b-c) Power losses in different voltages and frequencies. In all cases, the loss in is almost 2.3 times larger than that in NV 6110. GaN transistors tested exhibit power losses that increase with voltage, switching frequency, temperature and dc current. Whereas the case of dc current is explained by conduction losses, the exceedingly high loss levels that increase with voltage and frequency calls for further investigation. GaN transistor losses are comparable or larger than their SiC diode counterparts and are large enough to significantly degrade the rectifier efficiency at 100s of volts and 10s of megahertz operations. ACKNOWLEDGMENT We would like to thank Texas Instruments for funding this work through the Energy/Power Management Systems focus area of the Stanford SystemX Alliance. We would also like to thank Navitas Semiconductor for providing parts for evaluation. REFERENCES [1] L. Raymond, W. Liang, L. Gu, and J. Rivas, 13.56 mhz high voltage multi-level resonant dc-dc converter, in 2015 IEEE 16th Workshop on Control and Modeling for Power Electronics (COMPEL), July 2015, pp. 1 8. [2] S. H. Jayaram, Sterilization of liquid foods by pulsed electric fields, IEEE Electrical Insulation Magazine, vol. 16, no. 6, pp. 17 25, Nov 2000. [3] F. Guo, X.-H. Ji, K. Liu, R.-X. He, L.-B. Zhao, Z.-X. Guo, W. Liu, S.-S. Guo, and X.-Z. Zhao, Droplet electric separator microfluidic device for cell sorting, Applied Physics Letters, vol. 96, no. 19, 2010. [Online]. Available: http://scitation.aip.org/content/aip/journal/apl/96/19/10.1063/1. 3360812 (c) [4] C.-H. Lin, J.-H. Wang, and L.-M. Fu, Improving the separation efficiency of dna biosamples in capillary electrophoresis microchips using high-voltage pulsed dc electric fields, Microfluidics and Nanofluidics, vol. 5, no. 3, pp. 403 410, 2008. [Online]. Available: http://dx.doi.org/10.1007/s10404-008-0259-7 [5] L. C. Raymond, W. Liang, and J. M. Rivas, Performance evaluation of diodes in 27.12 mhz class-d resonant rectifiers under high voltage and high slew rate conditions, in 2014 IEEE 15th Workshop on Control and Modeling for Power Electronics (COMPEL), June 2014, pp. 1 9. [6] Bottom-side cooled 650 V E-mode GaN transistor, GaN Systems, 2016, rev. 160302. [7] GaN Power Low-loss Switch, Transphorm, Dec. 2014. [8] L. Raymond, W. Liang, J. Choi, and J. Rivas, 27.12 mhz large voltage gain resonant converter with low voltage stress, in 2013 IEEE Energy Conversion Congress and Exposition. IEEE, 2013, pp. 1814 1821. [9] G. Systems, 650v enhancement mode gan transistor, datasheet, September 2015. APPENDIX A We provide the voltage and current ratings of GaN transistors and SiC diodes evaluated in this paper. Shown in Fig. 7 are maximum allowable blocking voltages and continuous currents at case temperature of 100-135 C according to the datasheets. We made our best effort to select parts with similar ratings for the fairest possible comparison. 10A 8A 6A 4A 2A STPSC406 C3D04060E C3D1P7060Q CSD04060E 0A 0V 200V 400V 600V 800V Max. Blocking Voltage Fig. 7: Comparison of maximum blocking voltages and continuous drain currents of GaN transistors and SiC diodes evaluated in this paper. Blue markers and red markers correspond to GaN transistors and SiC diodes, respectively. APPENDIX B We estimate the impact of inductor losses on GaN transistor loss measurements. Thermal imaging throughout the experiment reveals that inductor losses are dominant source of power dissipation among passive components in the rectifier circuit. Therefore, we focus on measuring the thermal resistance between inductor soldering pads and GaN transistors. As shown in Fig. 8a, we attach a 10 Ω resistor to the soldering pad of the resonant inductor L res and the impedance matching inductor L. We deliver constant power to one of the resistors for around two minutes until the system reaches thermal equilibrium. We then record the average case temperature of the GaN transistor as in Fig. 8b. The thermal resistance between the inductor and GaN transistors is 0.38 C/W for both L res and L. This thermal resistance translates to +5 % error or less in the GaN transistor loss measurement. This estimate is based on several conservative assumptions: first, the thermal conductivity between the

inductor and the circuit board is as high as that in the case of surface-mount resistors in Fig. 8a; second, the rectifier operation time is several minutes for thermal equilibrium, which has never been the case during the transistor loss measurement; lastly, the inductor loss and the burst duty cycle are simultaneously at their maximum (6 W and 0.33, respectively), which is the opposite of the usual condition where those two variables are negatively correlated. Resistor in place of impedance matching L Resistor in place of Lres Dgnd Fig. 8: Experimental setup to measure the thermal resistance between inductor soldering pads and GaN transistors. We attach a 10 Ω resistor to the soldering pad of the resonant inductor Lres and the impedance matching inductor. We deliver constant power to the resistors and measure the average temperature rise in the GaN transistor surface by thermal imaging.