IEEE PEDS 2017, Honolulu, USA 12-15 December 2017 Application of GaN Device to MHz Operating Grid-Tied Inverter Using Discontinuous Current Mode for Compact and Efficient Power Conversion Daichi Yamanodera University of Tsukuba Graduate School of Pure and Applied Sciences Ibaraki 305 8573, Japan Takanori Isobe, Hiroshi Tadano University of Tsukuba Faculty of Pure and Applied Sciences Ibaraki 305 8573, Japan Abstract This paper studies on a grid-connecting inverter using a gallium nitride (GaN) device aiming for passive components size reduction by very high switching frequency operation. This paper proposes to apply a discontinuous current mode (DCM), which does not require dead-time and current feedback control, which are usually required for a continuous current mode (CCM) operation. These features enable a good modulation performance with a MHz-class high switching frequency operation without difficulties coming from the very high switching frequency. This paper reports experimental demonstrations of the DCM gridconnecting inverter using GaN- high electron mobility transistors (GaN-HEMT) with 1 MHz carrier frequency, and discusses output current harmonics and losses. I. INTRODUCTION Conventionally, high efficiency and downsizing of grid-tied inverters are desired. The most popular approach to achieve downsizing is to increase switching frequency so that the magnetic components can be small. In recent years, high efficiency and downsizing of the grid-tied inverter are expected to be achieved by using Gallium Nitride (GaN) devices with a very high carrier frequency [1]. However, the conventional grid-tied inverter using a continuous current mode (CCM) requires a dead-time and it will cause a large voltage error in the very high switching frequency and therefore output current distortion. It should also be mentioned that the inductance of the grid-connecting inductor can be reduced by increasing the switching frequency; however, at the same time its percent impedance in the line frequency will also be very low; therefore, the current control against to distorted grid voltage and/or dead-time voltage could be highly difficult. Moreover, the conventional inverter also has problems related to switching losses, and those will highly impact on efficiency in the very high frequency operation. To address those problems mentioned above, this paper proposes to apply a discontinuous current mode (DCM) operation [2] to the grid-tied inverter. In the DCM operation, the switches are not operated complementary, but one device in a leg is controlled for a half line cycle; therefore, the deadtime is not needed. The current control scheme of the DCM is completely different from that for the CCM. Actually the 978-1-5090-2364-6/17/$31.00 c 2017 IEEE current control is not needed for the DCM and only an openloop duty ratio control is applied. The current is still affected by the voltage distortion; however, the impact is comparatively low. In addition, the turn-on is performed with zero current; therefore, turn-on switching loss can be reduced. This paper reports experimental demonstration of a 1 MHz operated grid-connecting inverter using a GaN- high electron mobility transistors (GaN-HEMT) device and the discontinuous current mode operation. The device characteristics of the GaN-HEMT device used for the demonstration is discussed in section II, the operation principles of the DCM is introduced in section III, and the demonstration results are reported in section IV. II. CHARACTERISTICS OF GAN DEVICE A GaN-HEMT device (GS66504B from GaN systems: 650V 15A) is selected to be discussed to apply the grid-tied inverter for enabling MHz operation, since the GaN-HEMT has comparatively low parasitic capacitance, C iss, C oss, C rss, in comparison with GaN-GIT and GaN-Cascode devices for the same voltage ratings; therefore, can be expected to achieve higher switching frequency operation. Its main parameters are listed in Table I. The GaN-HEMT is packed in a distinctive package named GaN px, which has comparatively low parasitic inductance inside of the package; therefore, has advantages for MHz operation. The switching characteristics of the GaN-HEMT device were tested with the boost converter configuration as shown in Fig. 1 and Table II shows measurement condition. Fig. 2 shows measured switching waveforms at turn-on and turn-off. It could be observed that the turn-on time, t on, was 4.9 ns and the turn-off time, t off, was 3.0 ns with this current (approx. half of the rating). As a result, it was confirmed that the rate of the time required for switching is 0.8% of one switching cycle of 1 MHz operation. This indicates that the GaN-HEMT device is sufficiently capable of operating at 1 MHz. It can be concluded from the results that the GaN-HEMT device has enough high switching speed. 1,003
TABLE I MAIN DEVICE PARAMETERS OF GS66504B Rated Voltage V ds 650 V Rated Drain Current I d 15 A On Resistance R on 130 mω Input Capacitance C iss 130 pf Output Capacitance C oss 33 pf Reverse Capacitance C rss 1 pf SW H V gsh L i swh C V out SW L V in v ds v gsl i ds Fig. 1. Schematic circuit diagram for the test of the switching characteristics. A spike current in i ds was observed at turn-on, which seems to be the current charging C oss of the other side device turning off. The observed dv/dt at turn-on was -32.7 V/ns and C oss of the device is around 40 to 80 pf; consequently, spike current around 1.0 to 3.0 A could flow. The spike current could increase the turn-on loss since the area produced by i ds and v ds corresponds to the turn-on loss. III. GRID-TIED INVERTER OPERATED IN DCM A. Operation Principles Fig. 3 shows the schematic circuit diagram of a single-phase grid-tied inverter which can operate in the DCM operation. It consists of a usual single-phase full-bridge inverter, an inductor whose inductance, L ac, is comparatively low, and a LC filter. Fig. 4 shows schematic waveforms. Due to the comparatively low inductance, the current in the inductor, i Lac, is discontinuous and series of triangular waveforms. The LC filter is essentially needed to smooth the waveform, and the grid current, i ac, becomes a continuous sinusoidal waveform. Fig. 5 and Fig. 6 show the possible current paths and schematic waveforms in a switching cycle in the positive grid current phase, respectively. One difference from the usual CCM operation is that either switch of a leg is controlled TABLE II PARAMETERS OF THE BOOST CONVERTER FOR TESTING DEVICE Input Voltage V in 100 V Output Voltage V out 200 V Gate voltage v gs -3 to 6V Inductance L 5 µh Capacitance C 0.8 µf Fig. 2. Observed switching waveforms of the GaN-HEMT device. Turn-on. Turn-off. for a half line cycle and the other is kept at off-state. i Lac increases by turn the devices U and Y on as shown as Mode 1; then, i Lac decreases with free-wheeling conduction, which is also available with the GaN device, as shown as Mode 2. Another difference is the zero current period as shown as Mode 3, which is achieved after the current becomes zero in Mode 2. During this mode, the output voltage of the leg is indeterminate in ideal circuit; however, oscillating by the resonance between the parasitic capacitance of devices and the inductance in actual implementation. A modulation technique is proposed in [2]. The average of i Lac in a switching cycle is almost equal to i ac ; therefore, the average or area of the triangular waveform should be controlled to be equal to the current set-point, i ac ; for instance, a sinusoidal current set-point as i ac = 2I ac sin θ, (1) where I ac is the current set-point in r.m.s., can be given. For this case, the duty ratio, d, to achieve the desired current can be expressed as d= 2Lac f sw (V dc + 2V ac sin(θ φ) ) sin(θ φ) I ac V dc (V dc, 2V ac sin(θ φ) ) (2) where f sw is the switching frequency, θ is the phase angle of the line voltage, φ is the angle difference of the current to the line voltage (simply power factor angle). From the equation, it can be noticed that I ac can be separated; therefore, d for 1,004
Fig. 3. Schematic circuit diagram of the single-phase grid-tied inverter for the operation in DCM. Fig. 4. Schematic switching pattern and resulting waveforms of the DCM operation. any current amplitude can be easily calculated by an off-line calculated look-up-table. B. Advantages of DCM for MHz Operation with GaN-HEMT This paper proposes that the DCM operation is suitable for MHz operation of the grid-tied inverter using the GaN- HEMT device. As mentioned above, the dead-time is not needed and switching loss is mitigated by applying the DCM operation. Those characteristics are advantageous to achieve the MHz operation in the grid-tied inverter. On the other Mode 1 Mode 2 (c) Mode 3 Fig. 5. Possible current paths of the DCM operated grid-tied inverter in positive output current phase. Fig. 6. Schematic waveforms of the DCM operation in positive grid current phase in the scale of switching cycle. hand, the DCM operation shown in this paper uses reverse conduction characteristic; thus, its forward voltage V f can be a disadvantage in some types of devices. However, the V f of the GaN-HEMT device is relatively low by an appropriate negative bias voltage applied to the gate when it is turn off. The completely different current control scheme of the DCM can also be advantage for the MHz operation. The reduced inductance of the grid-tied inductor achieved by the increased carrier frequency introduces difficulty in the current control. The CCM operated grid-tied inverter can be modeled as two voltage sources connected via an inductor; therefore, the low inductance will increase the current distortion caused by voltage distortions. The dead-time also introduces the voltage distortion. The proposed DCM control has an attractive characteristics that the triangular current waveform of i Lac always starts from zero and reset every switching cycle and its average in a switching cycle will be linear to the voltages; therefore, current control using a current sensor and feedback is not needed. These feature brings possibility to reduce the grid connecting inductance further by increasing switching frequency. C. Switching Characteristics with DCM Operation Switching characteristics with the CCM and DCM operations were tested and compared with the same boost converter configuration shown in Fig. 1 and the same setting as listed in Table II. The resulting waveforms with the CCM are shown in Fig. 7, and ones with the DCM are shown in Fig. 7. The average input current in both cases were set at approximately 5 A. The spike current at turn-on was observed in both cases, but in the DCM, it was observed with almost zero current flowing into the inductor since it was operated in zero current switching; therefore, contribution to the turn-on loss was reduced. The amplitude of the spike current in the DCM was lower than one with the CCM. The reason for that can be thought as the reduced voltage charged in C oss of the other side device to be turned off. D. Design Circuit Parameter The DCM operated grid-tied inverter controls current by the duty ratio as equation 2 to achieve the given current set-point, I ac. The output current, i ac, is equal to the average value of 1,005
TABLE III CIRCUIT PARAMETER OF THE FABRICATED CONVERTER AC line Voltage V ac 100 V Line frequency f ac 50 Hz DC voltage V dc 180 V Rated AC current I ac 3.0 A Inductor L ac 2.4 µh Switching frequency f sw 1 MHz Filter inductor L f 5.0 µh Filter capacitor C f 390 nf Fig. 7. Measured waveforms with the CCM and DCM operation around turn-on with CCM operation, DCM operation. the inductor current, i Lac, in a switching cycle; therefore, the instantaneous output current can be expressed as i ac = 2I ac sin(θ φ) = V dc (V dc 2V ac sin(θ φ)) L ac f sw (V dc + 2V ac sin(θ φ)) d2. (3) On the other hands, the maximum duty ratio, d, to operate the converter within the DCM can be derived as d max = V dc + 2V ac sin(θ φ) 2V dc. (4) From equations 3 and 4, the maximum instantaneous current with the DCM can be derived as i max = V dc 2V ac 2 sin 2 (θ φ) 4L ac f sw V dc. (5) For instance, the maximum output current in rms, I ac.max, with unity power factor as φ = 0 can be expressed as I ac.max(φ=0) = V dc 2 2V ac 2 4 2L ac f sw V dc. (6) To achieve a given rated power with given input and grid voltages, L ac and f sw can be design parameters. Once the switching frequency is fixed, the maximum inductance to achieve the rated power within the DCM is determined. At the same time, to reduce the peak current in the inductor and semiconductors, L ac should be high as possible within the range. IV. EXPERIMENTAL DEMONSTRATION A. Experimental Setup To verify the proposed concept and operation principles, a laboratory prototype as shown in Fig. 8 was fabricated. The circuit parameters are listed in Table III. As switching devices, the GaN-HEMT (GS66504B 650V 15A) discussed in the previous section were used, and evaluation boards (GS66504B-EVBDB) including two GaN devices and gate drives provided by the manufacture were used to buid the full bridge circuit. A grid connecting inductor referred as L ac was fabricated as shown in Fig. 8. The grid connecting inductor consists of a troidal air-core and Litz wires to avoid possible high magnetic core loss generated by the high switching frequency, and to mitigate skin effect caused by the discontinuous current waveform. The Litz wire configuration was φ0.08 30 for L ac. The grid connecting inductor has two windings providing its half inductance for both lines to make a balance. The grid connecting inductor was designed to achieve a maximum power of 500 W with f sw = 1 MHz, V ac = 100 V, that is considering some margin to ensure the DCM operation at the rated power of 300 W. Fig. 9 shows measured inductance of the inductor and its equivalent series resistance. The self resonance frequency seems to be higher than 8 MHz, which is the upper limit of the LCR meter. As a controller, a DSP based controller using TMS320F28377S from Texas Instruments was used. To ensure a high resolution for modulation, a high resolution PWM modulator block provided by the microcontroller was used. A simple control scheme based on the operation principle as shown in Fig. 10 was implemented on the controller. A single-phase PLL using a Second Order Generalized Integrator (SOGI) [3] was implemented, and gate signals with duty ratio corresponding to the line phase angle provided by the PLL were generated. B. Evaluation of Operation and Waveforms Experimental evaluations in the range between 50 W to 300 W were performed with the fabricated converter. Measured output current waveform in line cycle scale and detail waveforms including device voltage, inductor current in switching cycle scale are shown in Fig. 11. As can be seen from Fig. 11, an in-phase sinusoidal current with the grid voltage was observed; therefore, the proposed control scheme 1,006
1 UY + 0 1MHz PLL Voltage Sensor duty ratio d eq.(1) VX Gate Driver DSP Control Board Fig. 10. Schematic control block diagram for the DCM operated converter used in the experiments. Fig. 8. Overview of the fabricated converter. Semiconductor part consists of two evaluation boards. Fabricated grid connecting inductor with a troidal air-core. Fig. 9. Measured inductance, Lac, and the equivalent series resistance, Rs, of the grid connecting inductor. and the PLL were verified to operate correctly. An 1 MHz DCM operation can be seen from Fig. 11, and the zero current switching (ZCS) was confirmed as that ilac has zero current period and increased from the zero level in every switching cycle. Also the natural turn-off, in which ilac decreased to zero without forced turn-off, was confirmed. The harmonic analysis on the output current iac was performed and the result is shown in Fig. 12. Total harmonic distortion (THD) considering under 49th order was 6.94%. From Fig. 11, some differences from ideal switching waveforms could be seen. One is a current oscillation observed during the period corresponding to Mode 3. It may cause an nonuniform turn-on in a line cycle and a variation of the peak and average currents depends on turn-on timing, and may cause output current distortion [4]. For this problem, a modified DCM operation to make an uniform zero current period to avoid the oscillation affecting output current distortion has been proposed in [5], and can be applied to the proposed GaN inverter. As another difference from the ideal waveform, a negative current in ilac was observed after the current reached at zero. This is a part of the oscillation in Mode 3 and can be thought due to parasitic components. This negative current in ilac also affects on iac. It can be said that this component should be predicted and cancelled by modifying the duty ratio given in equation 2. C. Loss Analysis The input and output power were measured by a digital power meter WT-1800, and the total loss and efficiency were evaluated. The measured efficiency at 300 W was 89.6%. To analyze the loss contribution from components, a loss break-down based on waveforms obtained by the oscilloscope (HDO4034) was performed. GaN conduction losses of forward and reverse directions were calculated with measured instantaneous inductor current, ilac, and on resistance, Ron, and Vf characteristics obtained from the data sheet. For reverse current conduction, Vf characteristics corresponding to the given negative gate voltage were used. The inductor loss was calculated by assuming only winding resistance of the inductor conducting the discontinuous current since there is no magnetic cores were used and the current in the filter inductor does not include high frequency components so much. The winding loss of Lac was calculated by summing losses generated by harmonic components of the current and equivalent series resistance at the corresponding frequency. The equivalent series resistance, Rs, shown in Fig. 9 was used. An example of the harmonic current components is shown in Fig. 12. The switching frequency was 1 MHz; therefore, its integer multiples could be seen. For the loss calculation, those components but less than 8 MHz were considered. As can be seen from the figures, the MHz components can generate major winding loss due to the non negligible AC resistance in the range of frequency. The rest of the loss, which can be obtained by subtracting above identified loss components from the total loss, is referred 1,007
Voltage (V) 200 100 0 i Lac v ds i ac 10 0 Current (A) 0 1 2 time ( s) Fig. 11. Experimentally measured waveforms with the rated power operation. Grid voltage, v ac, and output current, i ac, in line cycle scale. The voltage applied to a device, v ds, the inductor current, i Lac, and the output current, i ac, in switching cycle view around peak grid voltage phase. Fig. 12. Magnitudes in r.m.s value of harmonic components with approximately rated operation (305 W) of Output current, i ac. Inductor current, i Lac. as other losses. This includes the loss generated by the filter, C oss shorting at turn-on and switching losses. The resulting loss break-downs are shown in Fig. 13. It can be seen from the figure that the reverse conduction loss was most highest loss component. The reverse voltage V f of the GaN device can varies with the applied negative gate voltage; however, the high enough gate voltage to ensure safe operation was applied in this laboratory prototype. The reverse gate voltage optimization or applying positive gate voltage for the device conducting the current like synchronous rectification technique should be considered to be applied. The second major loss component was the inductor winding loss due to the very high frequency operation and high ripple components in i Lac. The Litz wire configuration was not optimized for this prototype; therefore, much more thin wire could be used. This prototype does not use magnetic cores; however, the use of the magnetic core can decrease the number of turn therefore the winding loss can be reduced. The winding loss and core loss are trade off; therefore, design optimization for overall loss reduction should be applied to the inductor design. The overall system efficiency was around 90%, however, Fig. 13. Resulting total efficiencies and losses as function of the output power. Loss break-down results are also shown. this prototype was operated with comparatively low voltages than the device rated voltage. From the loss break-down results, it can be said that the system efficiency will be improved when the prototype operates with more high voltage; for instance 200 V grid voltage. V. CONCLUSION This paper proposed to apply the discontinuous current mode (DCM) to the single-phase grid-connecting inverter using GaN devices for MHz class operation. Experimental verification using a fabricated laboratory prototype with 1 MHz 1,008
switching frequency up to 300 W was conducted. A simple loss analysis was also performed. Drastically low inductance of the grid connecting inductor and filter inductor were confirmed to be used with the proposed DCM inverter. However, the experimental results indicated some points to be improved. For the observed output current waveform distortion, the proposed control scheme must be improved considering switching characteristics of the GaN device. To improve the efficiency, the reverse conduction losses of the GaN device need to be improved. That can be achieved by gate voltage optimization, or synchronous rectification. But the synchronous rectification will introduce an additional complexity to the control; therefore, feasibility to implement in this very high switching frequency should be investigated. REFERENCES [1] Y. Lei, C. Barth, S. Qin, W. Liu, I. Moon,A. Stillwii, D. Chou, T. Foulkes, Z. Ye, Z. Liao, A 2 kw, single-phase, 7-level, GaN inverter with an active energy buffer achieving 216 W/in 3 power density and 97.6 % peak efficiency, IEEE APEC, pp.1512 1519, 2016. [2] T. Isobe, K. Kato, N. Kojima, Soft-Switching single-phase gridconnecting converter using DCM operation and a turn-off snubber capacitor, IEEE Transactions on Power Electronics, Vol. 29, No. 6, pp.2922 2930, 2014. [3] M. Cibotaru, R. Teodorescu, F. Blaabjerg, A new single-phase PLL structure based on second order generalized integrator, IEEE Power Electronics Specialists Conference, 2006. [4] Koen De Gusseme, David M. Van de Sype, Alex P. M. Van den Bossche, and Jan A. Melkebeek, Input-Current Distortion of CCM Boost PFC Converters Operated in DCM, IEEE Transactions on Industrial Electronics, Vol. 54, No. 2, pp.858 865, April 2007. [5] J. Zhang, R. Barrera-Cardenas, T. Isobe, H. Tadano, Trapezium Current Mode (TPCM) Boundary Operation for Single Phase Grid-tied Inverter, 2017 IEEE Energy Conversion Congress and Exposition (ECCE 2017), Cincinnati, OH, United States, Oct 1 5, 2017. 1,009