An all SiC MOSFET High Performance PV Converter Cell
Dipankar De1, Alberto Castellazzi1, Adane Solomon1, Andrew Trentin1, Masataka Minami2, Takashi Hikihara2 1
UNIVERSITY OF NOTTINGHAM Power Electronics, Machines and Control (PEMC) Group Nottingham NG7 2RD, UK Tel.: +44/ (0) – 1159515568 Fax: +44 / (0) – 1159515616 E-Mail:
[email protected] URL: http://www.nottingham.ac.uk 2
KYOTO UNIVERSITY Laboratory of Advanced Electrical Systems Theory, Japan Tel.: +81(75) – 383-2237 Fax: +81(75) – 383-2238 E-Mail:
[email protected] URL: http://www.kyoto-u.ac.jp
Acknowledgements The authors gratefully acknowledge the appreciated contribution of Mr. Nobuhiro Hase and Dr. Takashi Nakamura of ROHM Semiconductor, Kyoto, Japan, for supplying components used in this study.
Keywords Silicon Carbide (SiC), MOSFET, Efficiency, Photovoltaic, Multilevel converters.
Abstract Recent studies have pointed out the benefits of using Silicon Carbide (SiC) devices in photo-voltaic power conversion. In Particular, SiC Power MOSFET technology has greatly advanced over the last years and has presently reached sufficient maturity to stimulate a concrete interest in the development of power conversion circuits based entirely on this technology, in view of the clear potential advantages it offers over alternative SiC device technologies (e.g., JFET, BJT). This paper presents a thorough characterization of an all SiC MOSFET based single-phase bi-directional switched neutralpoint-clamped (BSNPC) three level inverter, in which, for the first time, SiC Power MOSFETs of different voltage ratings (1200 and 600V) are used. A parametric experimental characterization of the power cell performance is carried out, separating the effects of output power, heat-sink temperature and switching frequency and load variations by means of bespoke heat-sink design. The effect of relying exclusively on the MOSFET body-diode for inductive load current freewheeling is critically assessed against usage of an external SiC Schottky diode. The experimental results are compared with a mixed approach design, where Silicon (Si) devices are used for the lower voltage switches and SiC MOSFETs are kept for the higher voltage ones, deriving a clear indication of the superior possibilities offered by SiC Power MOSFETs for improved efficiency, power density and reliability, key aspects of power electronics technology evolution.
Introduction SiC power devices are well known to offer a number of potential advantages over their Silicon (Si) counterparts (e.g., higher switching frequencies, higher operating temperature [1-4]). In particular, it is broadly accepted that main benefits are to be drawn at voltage levels indicatively above 600V. Recent studies have focused specifically on the benefits of SiC in photo-voltaic (PV) [5-8], demonstrating the potential to improve efficiency and achieve higher switching frequencies and power densities due to reduced filters and heat-sink sizes. However, most published work still refers to a hybrid combination of SiC and Si devices, with some studies focusing on SiC JFETs and BJTs [9]; no study has yet presented an all SiC MOSFET based power cell, neither a thorough characterization of performance over temperature (i.e., in most works, temperature varies with output load, since the heat-sink is designed based on constant cooling parameters). In this work, we make use of both 1200V and 600V rated SiC MOSFETs to develop a fully SiC based single phase bidirectional switch neutral point clamped (BSNPC) inverter [10-11], Fig. 1(a), particularly relevant to transformer-less PV applications [12]. This paper investigates the efficiency over a broad range of output power for different switching frequencies and different heat-sink temperatures. In contrast to most previous investigations, here, the heat sink temperature is kept constant when varying the load conditions by controlling the amount of active cooling, so as to clearly separate the influence of temperature on the system performance. Also, an explicit comparison is carried out of the performance when using the MOSFETs body-diode as a freewheeling element for the load current and when an external anti-parallel SiC Schottky diode is introduced. Benchmarking with state-of-art Si IGBT with external anti-parallel SiC Schottky diode is performed and a critical discussion is proposed.
Description of the Bi-Directional Switched Neutral Point Clamped Inverter Among various transformer-less topologies for PV application, BSNPC converter is one of the interesting candidate. This topology is chosen to compare the influence of 600V SiC devices mainly as bi-direction device on the overall converter efficiency. The power circuit is shown in Fig. 1(a) for grid connected applications; the design values and test conditions applied in the present work are summarized in Table I. Vdc represents a DC source (e.g., solar panels output). Transistors M1 and M4 constitute a half-bridge switch and are rated for Vdc; M2 and M3 implement a bi-directional switch configuration and are rated for Vdc/2. Here, 600V devices are used to implement the bi-directional switch, while 1200V transistors are used for the half-bridge switch. Use of external anti-parallel diodes is optional in the case of Power MOSFETs, because of the presence of an intrinsic body diode between source and drain terminals. An output filter is needed to smooth the load current ripple. In the laboratory, an RL load is used for testing, as indicated in Fig. 1(b). In this study, the power rating of the converter is 2.5kW, which is a representative figure for residential PV application [8]. A fixed dead-time of 400ns was chosen as the best compromise between low distortion of the output current waveform at high switching frequencies (up to 64 kHz; the inverter is tested in open-loop in this work) and accommodating operation with both SiC and Si devices with a unified gate-driver design. The bidirectional devices (M2 and M3) are circled to highlight the focus of the benchmark exercise carried out in this work. The modulation scheme is after [10]; to generate the required gate signals for all the switches, the modulation strategy illustrated in Fig. 2 is implemented: two level shifted carrier signals are compared with the modulating signal to generate the switching pulses for M1, M4 and M2, M3. In Fig. 2, which also shows the resulting drive signal for each MOSFET and the resulting pole voltage with respect to the DC bus mid-point (VPO in Fig. 1 (b)), the carrier frequency is intentionally kept low for the sake of clarity.
Fig. 1: (a) Grid connected PV inverter, (b) Simplified system with RL load. Fig. 3 shows the pole voltage and corresponding sinusoidal load current for a switching frequency fs = 16 kHz. In this topology, the diodes in anti-parallel to the MOSFETs in the half-bridge switch (M1, M4) only conduct current for a short time, in correspondence of the sign reversal of the load current due to near unity factor loading condition (encircled in Fig. 3(c)); on the other hand, the diodes in the bi-directional switch (M2, M3) are conducting whenever M1 and M4 are off (see Fig. 2 and Fig. 3(d)). This observation is important for the discussion in the next section about the introduction of external anti-parallel SiC Schottky diodes. In the application, increasing the switching frequency enables a consistent reduction of the size of the passives used in the filter, hence the interest for switching at the highest possible values. Typically, a trade-off is needed with the increase in switching power losses.
Table I: Power Cell Design Parameters and Test Conditions Pmax
2.5kW
Vdc
650V
L
1.5mH
C
2500μF
Rb
22kΩ
fs
16 – 64kHz
Dead-time
400ns
Fig. 2: Illustration of switching pulse generation logic and resulting pole voltage.
Fig. 3: Simulation results: (a) Inverter output voltage (VPO), (b) load current, (c) body-diode current of MOSFET M4, (d) body-diode current of MOSFET M2.
Experimental Characterization All SiC inverter Fig. 4 shows a photograph of the experimental power converter. The gate driver board is kept very close to the power cell to minimize the noise pick up at the gate signals. Here, the power cell performance is tested in open-loop. As this particular type of converter is based on centre-tapped dclink capacitor, it requires large capacitance values to minimize the low frequency voltage oscillation at the capacitor mid-point. Both electrolytic and high frequency capacitors are used to suppress the voltage oscillation at the DC-bus. The carrier and modulating signals are derived from external signal
generators and are duly processed on the gate-driver to deliver the required switching sequence. The drive signals are passed through isolation, dead-time and current-boosting circuitry to prevent shoot through and enable high-frequency operation of the power transistors. Initially the power cell was implemented with only 4 MOSFETs (i.e., no external free-wheeling diodes). Gate driver voltage levels are kept +20V to -3.3V for SiC MOSFETs. The experimental waveforms of the pole voltage and load current are shown in Fig. 5 for a switching frequency of 16 kHz, confirming excellent agreement with the design intentions (simulation results of Fig. 3). The semiconductor devices used for the implementation and comparison are listed in Table. II in Appendix.
Fig. 4: Photograph of the experimental setup (M1, M4: TO247 case and M2, M3: TO220 case), C_l: Electrolytic capacitor, C_h: high frequency low ESR capacitor, R_b: bleeder resistor.
Fig. 5: Measured Results: Inverter output voltage (VPO) and load currents The heat-sink was equipped with a thermocouple on the device side, in the center relative to all devices, so as to get an estimate of the average heat-sink and device temperature. Fig. 6 reports the measured efficiency under various conditions. Some interesting observations can be made about these results: - In Fig. 6 (a), where the heat-sink temperature is 50ºC, one can see that increasing the switching frequency by as much as a factor 4 only implies a reduction in efficiency of less than 1% at an output load of 2.5 kW (less than 25 W difference in total power dissipation). This is clearly a very interesting
result, when one considers the reduction in the value and size of the output inductor achievable by increasing the switching frequency by 4. - In Fig. 6 (b), it can be seen that the above observation holds true even at higher values of heat-sink temperature. Moreover, the effect itself of increasing the heat-sink temperature, illustrated for an output load of 2.5 kW and for different values of the switching frequency, is in first approximation negligible over the relatively narrow range of temperatures considered here, causing a decrease in efficiency of about 0.3% (7.5 W) in the worst case. Fig. 6 (c) proposes an alternative summary of the same results, confirming the excellent performance stability of SiC Power MOSFETs against changes in temperature, opening up new possibilities for increased system level power density.
Fig. 6: Measured Efficiency as a function of: (a) output power, at a fixed heat sink temperature of 50ºC, for three different values of the switching frequency; (b) switching frequency at a fixed output power of 2.5 kW for three different heat-sink temperature values; (c) heat-sink temperature at a fixed output power of 2.5 kW for three different values of the switching frequency. In a second round of tests, the inclusion of external SiC Schottky diodes as free-wheeling elements was considered to assess its impact on the converter efficiency. In view of the observations made above (Fig. 3 (c) and (d)) regarding the current conduction of the body diodes of all the MOSFETs in the circuit, if a gain in efficiency is to be achieved, that will be due to the diodes in the bi-directional switch. Hence, 600 V Schottky diodes were connected in anti-parallel to M2, M3: the forward voltage drop of the Schottky diode was 1.4 V; whereas that of MOSFET body diode was 2.9 V (the diode forward voltage drop measurements were carried out at a bias current of 4A). It should be noted, that, in the case of SiC Power MOSFETs, the anti-parallel or body-diode only freewheels during the deadtimes, whereas current is conducted in the MOS channel afterwards, so no major differences in efficiency are to be expected. Indeed, Fig. 7 (a) shows the efficiency for different values of the switching frequency, at 2.5 kW output load and 70oC heat sink temperature; Fig. 7 (b) shows the efficiency for different values of temperature still at 2.5 kW output load and 32 kHz switching frequency with and without Schottky diodes. At the considered power levels, the improvement in efficiency is contained and might not really justify the addition of an external SiC diode in the circuit, both from a cost optimization and a circuit complexity point of view, provided that previously reported issues with the body-diode stability are overcome, as seems to be the case in new generation devices [13]: indeed, the first available SiC Power MOSFETs were always packaged with anti-parallel external Schottky body-diode, whereas nowadays they are mainly packaged stand-alone. In this work, notwithstanding prolonged testing, no indication of performance degradation was observed when
relying on the body-diode for current freewheeling. However it is interesting to note that the device heat concentration is slightly more with only SiC MOSFET case compared to SiC MOSFET + SiC diode solution. In particular, it was confirmed that the addition of an external Schottky diode does not influence the switching transitions. Fig. 8 shows the turn-on voltage and current waveforms for M4, in the case without, (a), and with, (b), external Schottky diode. Before the transition the current is flowing through bidirectional device (MOSFET M3 and body diode of M2) and the voltages across M1 and M4 are Vdc/2. In both cases, the transition is completed in less than 100ns and the current profile exhibits the same initial current peak. At the same time the current through one of the bidirectional device M2 (Fig. 9) does not exhibit any peak. This indicates that, as expected, the current peak is not due to any diode reverse recovery effect in either case, but rather is a capacitive current due to the charge up of M1 from Vdc/2 to Vdc.
Fig. 7: Efficiency comparison between 600V SiC MOSFET and 600V SiC diode + SiC MOSFET as bidirectional device (a) at 70oC and 2.5 kW loading (b) at 32 kHz switching frequency and 2.5 kW loading
Fig. 8: Voltage across and current through M4 at turn-on: (a) all SiC MOSFET case (b) SiC diodes connected anti-parallel to the bidirectional devices
Fig. 9: Voltage across and current through M2 at turn-on of M4
Mixed Si and SiC inverter To benchmark the results obtained above, a different configuration was tried out, where only the active devices in the bi-directional switch were replaced with Si IGBTs (see Table. II). Use of IGBTs necessarily requires the presence of an external anti-parallel free-wheeling diode due to the transistor unidirectional current conduction and reverse voltage blocking characteristics. The IGBT characteristics also imply that each anti-parallel diode must conduct current not only during the deadtimes, as in the case of the SiC MOSFETs, but over the whole conduction time of the bi-directional switch. For that reason a diode with higher current rating as in the case of the SiC MOSFETs is needed (Table. II). Experimental results for this configuration are compared directly against those obtained with a fully SiC inverter. Fig. 10 shows the comparison between fully SiC and Mixed SiC/Si converter efficiencies for various test conditions. In Fig. 10 (a) the heat-sink temperature is set an average representative value of 65 oC and the switching frequency is 16 kHz: although the difference is minor, a decrease in efficiency can be observed when the IGBT is used in the bi-directional switch. As evident from Fig. 10 (b), the decrease becomes more pronounced as the switching frequency is increased to 32 kHz. A switching frequency of 64 kHz was simply impractical with the chosen IGBT: tests revealed a major loss of performance due to non-complete exhaustion of the turn-off charge extraction at the subsequent turn-on (i.e., turn-off current tail still non zero).
(a)
(b)
Fig. 10: Comparison of Full SiC and Mixed Si/SiC inverter efficiency as a function of output load at a representative heat-sink temperature of 65oC and for two values of the switching frequency: 16 kHz (a) and 32 kHz (b).
(a)
(b)
Fig. 11: Comparison of Full SiC and Mixed Si/SiC inverter efficiency as a function of temperature at a representative output power of 2.5 kW and for two different values of switching frequency: 16 kHz (a) and 32 kHz (b). Fig. 11 proposes a comparison of the efficiency as a function of heat-sink temperature, again at 16 kHz (a) and 32 kHz (b). In this case, too, use of an IGBT implies a more pronounced loss of performance for higher temperatures, in particular at higher frequencies, which are on the other hand of interest for reducing the size of the cooling device.
Conclusion The results presented highlight important features of SiC Power MOSFETs in PV inverters. In particular, their stable performance against switching frequency and temperature enables high flexibility in overall system design, trading off between efficiency and power density requirements and goals. The results of the benchmarking exercise between an all SiC and a mixed Si/SiC solution clearly indicate that use of a mixed SiC and Si solution, with IGBTs and Schottky diodes as the bidirectional switch, is competitive in performance to that of a fully SiC Power MOSFETs based one only at relatively low heat-sink temperatures and switching frequencies, with severe limitations on design flexibility and optimization at system level. Also, the results point out interest in the availability of SiC power MOSFETs in voltage classes lower than 1.2kV, too, for the bespoke needs of BSNPC topologies. Indeed, here, if reference is made to a comparison of 600V IGBT with Schottky diodes against 600V SiC MOSFETs alone, the additional cost burden due to SiC as opposed to Si can be realistically estimated to be a factor 2 to 3 (keeping in mind that IGBTs require a higher current rating diode, as discussed above) for the implementation of the bi-directional switch. Such cost difference is readily compensated mainly by an increase of the switching frequency which enables a significant reduction of the output filter inductor value (although a higher switching frequency also has a beneficial effect on the size of the high frequency input capacitors, the overall impact of these components on system size and cost is negligible) and, partly, by a reduction of the cooling requirements. When using SiC Power MOSFETs, no additional diodes are strictly needed, so the cell implementation is simpler. Problems with body-diode stability due to stacking faults reported in past literature seem to have largely been overcome and no indication of degradation was observed when relying on the MOSFET body-diode. Again, from a cost perspective, with PV promising to undergo a large market booming expansion worldwide, the possibility offered by SiC MOSFETs to keep comparable levels of performance with and without an external anti-parallel Schottky diode is also definitely of interest, as a reduced number of components also typically affects the system reliability positively. Other possible potential benchmark options, for instance, using super-junction MOSFETs rather than IGBTs in for active devices in the bi-directional switch, have presently been neglected due to known issues with high temperature coefficient of on-state resistance, gate-charge requirements and output capacitance energy storage, which impair the realization of fast (i.e., high dV/dt), hence low dissipative commutations of all devices in the circuit.
References [1] Majumdar G.: Recent technologies and trends of power devices, IWPSD 2007, pp. 787-792 [2] Hudgins J. L., Simin G. S., Santi E., and Khan M. A.: An assessment of wide band-gap semiconductors for power devices, IEEE Trans. Power Electron., vol. 18, no. 3, May 2003, pp. 907-914. [3] Funaki T., Balda J. C., Junghans J., Kashyap A. S., Mantooth H. A., Barlow F., Kimoto T., and Hikihara T.: Power conversion with SiC devices at extremely high ambient temperature, IEEE Trans. Power Electron., vol. 22, no. 4, July 2007, pp. 1321-1329. [4] Wood R. A., and Salem T.E.: Evaluation of a 1200V, 800A all SiC dual module, IEEE Trans. Power Electron., vol. 26, no. 9, Sept. 2011, pp. 2504-2511. [5] Ho C. N., Breuninger H., Pettersson S., Escobar G., Canales F.: A comparative performance study of an interleaved boost converter using commercial Si and SiC diodes for PV applications, IEEE Trans. Power Electron., vol. 28, no. 1, Jan 2013, pp. 289-299. [6] Araujo S. V., and Zacharias P.: Analysis on the potential of silicon carbide MOSFETs and other innovative semiconductor technologies in the photovoltaic branch, EPE 2009, pp. 1-10. [7] Araujo S. V., and Zacharias P.: Perspective of high voltage SiC-semiconductor in high power conversion systems for wind and photovoltaic sources, EPE 2011, pp. 1-10. [8] Mallwitz R.: SiC JFETs for photovoltaic Inverters – advantages and challenges, SiC and GaN User forum – Potential of wide band gap semiconductors in Power Electronics applications, 2011. [9] Guedon F., Singh S., McMahon R., and Udrea F.: Boost converter with SiC JEFTs: comparison with CoolMOS and tests at elevated case temperature, IEEE Trans. Power Electron., vol. 28, no. 4, Apr. 2013, pp. 1938-1945. [10] Kaminski B., Koczara W., and Al-Khayat N.: A three level inverter concept for low voltage applications, EPE 2007, pp. 1-8. [11] Ma L., Sun K., and Jin X.: A transformer-less method for conventional three phase full-bridge topology to conergy NPC topology, ICEMS, 2011, pp. 1-5. [12] Kerekes T., Teodorescu R., Rodríguez P., Vázquez G., and Aldabas E.: A New High-Efficiency SinglePhase Transformerless PV Inverter Topology, IEEE Trans. Ind. Electron., vol. 58, no. 1, Jan. 2011, pp. 184-191. [13] Agarwal A., Fatima H., Haney S., and Ryu S.: A New Degradation Mechanism in High-Voltage SiC Power MOSFETs, IEEE Electon. Device Letter, vol. 28, no. 7, Jul. 2007, pp. 587-589.
Appendix Table II: Description of the semiconductor devices used for comparison 600V SiC diode anti-parallel with 600V SiC MOSFET 600V SiC MOSFET 600V IGBT 600V SiC diode anti-parallel with 600V SiC IGBT 1200V SiC MOSFET
SCS110AG Not commercially available IGP20N60H3 C3D20060 CMF 20120
