1. Introduction
Power electronic systems and applications are considered to be one of the most essential parts of renewable energy sources (RES), electric vehicles (EV), EV chargers, and EV main inverters, reducing the environmental impact. High-capacity, high-temperature, and high-frequency power converters are increasingly demanded to increase power density, reduce costs, and save manpower. For this reason, the most significant features of new technology power converters are minimization of volume, maximization of efficiency, high reliability, and increased durability against short-circuit/overvoltage conditions. To achieve these goals, the transaction from silicon (Si) to wide band gap (WBG)-based semiconductor power switches is of great significance because of their outstanding features compared to Si ones.
Among WBG power switches, SiC MOSFET is considered to be the most promising alternative solution to conventional semiconductor devices in medium- and high-power-density power converter fields. This is attributed to its exceptional characteristics, such as the relatively mature technology, the low cost, and its more stable construction [
1,
2]. Indeed, the most significant features of SiC MOSFET are the high thermal conductivity and operating temperature capability, the higher breakdown voltage, its superior switching characteristics, the lower on-resistance, the usage of low complexity gate drivers, and the normally-off characteristic. Moreover, SiC MOSFET has no tail current leading to reduced switching losses as well as higher switching frequency [
1,
2,
3].
Nevertheless, SiC MOSFET offers less current capability than Si ones due to its smaller chip area. This derives from the lower maturity of the manufacturing process of SiC MOSFETs compared to Si ones which includes the lower yield in the wafer as well as the high thermal and mechanical stress in the device. As a result, the current ratings of commercially available discrete SiC MOSFETs with a maximum blocking voltage of 1.2 kV and 1.7 kV are within 120 A and 100 A, respectively. For this reason, the current rating is usually boosted by connecting multiple SiC MOSFET devices in parallel [
4,
5,
6,
7,
8].
However, the parallelization of SiC MOSFETs introduces the problem of current imbalance which is unpredictable. This results in uneven conduction and switching losses between parallel devices. This, in turn, causes uneven sharing of junction temperature, increasing the risk of SiC MOSFET(s) being led to thermal runaway [
9,
10]. Also, an over-current and, at the same time, overheating is quite possible. Therefore, it is essential to overcome any possibility of power device failure due to high junction temperature swing by suppressing the current imbalance and retaining distributed heat between power devices as equally as possible [
4,
11,
12].
Current imbalance can be caused by a device package parameter mismatch. For instance, the variation in on-state resistance (
RDS-on) causes unequal current sharing during the steady-state, leading to static current imbalance. Moreover, any difference in threshold voltage (
Vth) and trans-conductance (
gm) leads to uneven current distribution during turn-on and -off intervals, causing dynamic current imbalance [
4]. Puschkarsky et al., in [
13], experimentally proved the
Vth instability of high-voltage SiC MOSFETs which may be either short-time or even permanent. Asymmetry of the PCB layout of the power circuit may also affect the current sharing of power devices [
14]. Also, both types of imbalances result in unequal temperature rises and electromagnetic interference (EMI), endangering system reliability.
Over the last decade, the current imbalance issue has been addressed by many researchers, proposing techniques and methods capable of minimizing one or both current imbalance types between discrete parallel-connected SiC MOSFETs.
Refs. [
4,
15,
16] proposed several active current balancing methods that make use of current sensors to actively detect current imbalance. Subsequently, an analog controller receives the dynamic imbalance and suppresses it through a gate driver by matching the switching behaviors of the parallel switches.
Most proposed methods suppress static or/and dynamic current imbalance by using passive elements. Refs. [
11,
17] mitigated dynamic imbalance by adding extra coupled inductance and external same-size gate resistors. As a result, control voltages of parallel devices vary only during transient stages, eliminating the entire transient imbalance. In the same way, ref. [
18] eliminated dynamic imbalance with the addition of different-size gate resistors differentiating each gate loop impedance and eliminating turn-on dynamic imbalance. Ref. [
19] also deals with turn-on dynamic imbalance by making the gate resistor of the power switch with the smallest
Vth greater, delaying the charging process of its input capacitance (
Ciss). In refs. [
8,
20], static current imbalance was suppressed by adding same-size resistors, serially connected with the drains of the parallel-connected devices. In addition, ref. [
8] mitigates both types of imbalances with a differential mode choke. Similarly, refs. [
21,
22] suppressed the overall current imbalance with the incorporation of a series-connected coupled inductor with the drains of power devices. On the other hand, ref. [
23] suggests an alternative way to implement the two aforementioned methods to suppress current imbalance and avoid the disadvantages provoked by the usage of a coupled inductor. Finally, ref. [
24] eliminates the entire imbalance by connecting a planar transformer in series with the drains of each power device.
Refs. [
1,
25,
26,
27,
28,
29,
30] proposed novel screening methods for discovering SiC MOSFETs with very close technical parameters, such as
RDS-on,
Vth, and
gm, achieving a balanced current distribution without the requirement of a current imbalance suppression technique. Based on [
10,
31], the current imbalance, caused by the asymmetrical PCB layout of the power circuit, can be mitigated by lowering the deviations between drain and common source parasitic inductances. Nonetheless, ref. [
32] deals with the imbalance, attributed to the asymmetry of the power circuit layout by incorporating a common mode choke to each parallel SiC MOSFET gate loop. This method holds that the current imbalance is limited as the choke mutual inductance becomes larger. Ref. [
33] achieved an optimal transient current sharing by reducing the gate resistance, weakening the effect of
Vth mismatch.
In addition, ref. [
34] proposes a gate driver that generates PWMs with different time delays of picoseconds, suppressing dynamic imbalance. Additionally, ref. [
35] balances the dynamic imbalance with a multi-stage gate driver with the ability to change the gate resistor during the transient stages. Finally, ref. [
36] proposes a gate driver capable of varying the gate voltage to mitigate transient current imbalance.
All the aforementioned methods have the ability to eliminate either static or/and dynamic current imbalance between parallel SiC MOSFETs. On the other hand, most techniques cannot be implemented unless the technical parameters of SiC MOSFETs are known. For this reason, screening processes are needed and conducted with the aid of power device analyzers/curve tracers [
20]. Also, ref. [
37] proposed a method to monitor the on-resistance of SiC MOSFET. This necessity can be an inhibiting factor for their application in the industry since screening is an extremely costly and time-consuming process [
4]. The implementation of the methods, proposed by refs. [
4,
15,
16], can be realized without any screening process since dynamic current imbalance is mitigated with a closed-loop method that actively monitors and suppresses transient imbalance. However, static current imbalance suppression is not addressed. As a result, none of these techniques can minimize the whole current imbalance without knowledge of the device parameter mismatch. Also, the validity of most methods has not been tested under the condition of an asymmetrical PCB layout.
In this paper, an innovative, active, and autonomous open-loop current balancing technique is proposed which addresses the imbalance issue without the necessity of knowing the technical parameters of the power devices. In addition, the proposed technique can eliminate static and dynamic current imbalances actively and irrelevantly of the cause. In
Section 2, an analysis to investigate potential strategies for current imbalance suppression is conducted. In
Section 3, an analytical description of the structure, functions, and design guidelines of the active current balancing technique is provided. In
Section 4, two experimental tests are conducted to verify its effectiveness and efficiency. In
Section 5, extension guidelines of the proposed current balancing technique are proposed. Finally, in
Section 6, the most important conclusions of this research are presented.
4. Test Platform and Experimental Results
4.1. Test Platform
In previous work, the effectiveness and durability of the proposed technique against current imbalance were tested through simulation tests eliminating current imbalance automatically [
40,
41]. To experimentally verify the effectiveness of the proposed current balancing technique, an experimental test platform is constructed. The structure of the test platform and the proposed current balancing system are depicted in
Figure 8.
Table 1 lists all the equipment used. The test platform is derived by a DC–DC buck converter with two SiC MOSFETs C2M0080120D (M1 and M2) connected in parallel. As a free-wheeling diode, SiC Schottky E4D20120D is used. The power converter supplies a resistive load while an
LC filter is used for smoothing the output voltage. The realization of the proposed method includes the digital controller, two current sensors, and two active gate driver circuits which are powered by a separate DC power supply. All the capabilities of the digital controller can be realized with an algorithm and executed with an FPGA (field programmable gate array). For this reason, the Nexys A7-100T FPGA trainer board is used which includes the FPGA Artix-7 offering a clock speed of 500 MHz. Also, an algorithm is written in the VHDL programming language to execute all the digital controller functions. Since two gate drivers are utilized, the FPGA generates eight PWMs for the control of the power devices and the forward converters. The FPGA algorithm utilizes 14 switches (SW), the 7-segment displays, and the pushbuttons of the FPGA board.
Table 2 mentions in detail the function of each FPGA switch. The control results of the switching frequency and duty cycle of the SiC MOSFETs as well as the control parameters are displayed in the FPGA 7-segment displays and controlled with the help of the pushbuttons. Finally, the measurement of each drain current is achieved with a surface mount resistor
Rsense of 100 mΩ and 1 W, connected in series with the source pin of each parallel power device.
The DC bus power supply (
VBUS) of the DC–DC buck converter is implemented using a three-phase full-bridge diode rectifier which is connected in parallel with two capacitors (1500 μF/550 V) to smooth the output voltage of the bridge. The DC bus power supply also consists of a protection system that includes a resistor of 3.3 kΩ/10 W. When the resistor is connected in parallel with the capacitors, it discharges them for safety purposes. As shown in
Figure 9, the operation of the entire platform is controlled by a relay (Rel) which is powered by a single-phase AC power supply of 230 V. Once the switch (SW) is closed and the AC supply is ON, the relay connects the three-phase power supply with the rectifier bridge through a normally open (NO) three-phase switch, and at the same time disconnects the discharging resistor from the capacitors through a normally closed (NC) single-phase switch. Otherwise, the relay disconnects the bridge with the three-phase AC supply and connects the discharging resistor with the capacitors. In addition, four electric fuses are used, three in the three-phase AC supply (F1, F2, and F3) and one in the single-phase AC supply (F4), offering overcurrent protection to the experimental platform. The constructed experimental test platform is shown in
Figure 10.
As for the test conditions, the buck converter operates under a DC bus voltage of 200 V with a frequency and duty cycle of 25 kHz and 25%, respectively, supplying a load of 10 Ω. The initial Vdr values for activating and deactivating each parallel SiC MOSFET are 20 V and −5 V, respectively. Finally, RG-on and RG-off are equal to 10 Ω while the gate-source voltages of Maux,on and Maux,off are set to 9 V. Vctl-M,on,i and Vctl-M,off,i are the output voltages of the forward converter of Maux,on and Maux,off, respectively.
The initial values of the activation and deactivation voltages were selected as 20 V and −5 V, respectively, recommended by the datasheet of the utilized SiC MOSFETs C2M0080120D. The manufacturing company (Wolfspeed) has constructed a gate driver (CGD15SG00D2) designated for the driving of that particular SiC MOSFET model. CGD15SG00D2 uses an isolated DC/DC converter to generate two output voltages of 20 V and −5 V while its output power is 2 W with an efficiency of 86%, meaning that the input power of the driver is 2.3 W. On the other hand, the input voltage (Vin) of our proposed gate driver is 17 V to supply all three forward converters and an isolated DC/DC converter that generates the deactivation voltage of 5 V which is always constant. In case the FPGA controls only the forward converter which generates the activation voltage at 20 V, the input current is measured to be 0.15 A. Therefore, the input power of the proposed active gate driver when it supplies constant activation and deactivation voltages is almost 2.5 W, which is very close to the input power of the commercial gate driver (CGD15SG00D2). In case the FPGA controls all forward converters and the output voltages that control auxiliary MOSFETs are 9 V, the overall input current and power of the active gate driver are 0.21 A and 3.6 W, respectively. Therefore, both forward converters and the auxiliary MOSFETs add only 1.1 W of power consumption compared to the previous case in which the proposed active gate driver works in a similar way to the commercial driver. In the worst-case scenario, when the activation voltage reaches 23 V for the suppression of static imbalance, the input current and power are 0.29 A and 4.9 W, respectively.
4.2. Current Sensing System Accuracy
The current measurement of SiC MOSFETs requires sensors of high bandwidth because of their fast-switching speed. Based on [
42], surface mount resistors can offer exceptional measurement accuracy since their bandwidth can be on the order of hundreds of megahertz. One of the most important factors affecting their bandwidth is parasitic inductance which is inevitable because of the magnetic field induced by the current conducting the sensor [
43]. For this reason, surface mount resistor size should be as low as possible to provide low parasitic inductance in the order of nano or even picohenry. Also,
Rsense should have low resistance without significantly affecting the current level and offering quite low power losses. On the other hand, one of the pulse current measurement methods that are strongly recommended for the current measurement of WBG devices is the coaxial shunt resistors which can offer MHz or even GHz measurement bandwidth.
To examine the measurement accuracy of the utilized current sensor, two experimental tests are performed, measuring the current of one SiC MOSFET. In the first test, the power switch current is measured with the surface mount resistor of 100 mΩ. In the process, the SiC MOSFET current is measured with the coaxial shunt resistor SDN-414-01 which offers a measurement bandwidth of 400 MHz. The voltage developed across the current sense resistor is illustrated to the oscilloscope with the aid of a BNC coaxial connector and a BNC coaxial cable of quite high bandwidth, as shown in
Table 1.
Figure 11 depicts and compares the current waveforms of both tests during the conduction stage as well as the turn-on and -off intervals, demonstrating the surface mount resistor measurement accuracy. Compared to the coaxial shunt resistor, the surface mount resistor only shows a measurement delay of around 5 ns during the current ringing stages. However, this time delay difference can be ignored because both waveforms are identical during the rising, conduction, and falling stages of the drain current. In conclusion, for the purposes of this research, the current sense resistor offers high enough measurement accuracy of the drain current of SiC MOSFET.
4.3. Experimental Test Results
In this subsection, the effectiveness of the proposed method is tested by performing two experimental tests. In the first test, a pair of devices is connected in parallel causing current imbalance during steady and dynamic stages which may be attributed to the variation in the technical parameters between SiC MOSFETs. PCB layout of the power circuit is designed to be symmetrical to minimize the length differences between PCB traces to exclude any current imbalance attributed to mismatched parasitic inductances. In the second test, another pair of devices is connected in parallel which originally shows an equal current sharing between power devices. However, the layout was designed to be asymmetrical by connecting the power devices with different gate, drain, and source pin lengths. This leads to static and dynamic current imbalance caused mainly by the mismatch of the drain and common source parasitic inductances. Ref. [
31] argues that the mismatch of gate parasitic inductances has an almost negligible effect on dynamic current sharing. The experimental results are further compared under different test conditions (a) without and (b) with the proposed current balancing technique.
Figure 12 depicts the drain-source voltage (
VDS), developed across power devices, during the conduction stage and the turn-on and -off transitions. The experimental results of the two tests are illustrated in
Figure 13,
Figure 14,
Figure 15,
Figure 16,
Figure 17 and
Figure 18, depicting the drain currents of the parallel-connected SiC MOSFETs as well as their turn-on and -off gate currents and driving signals.
Table 3 and
Table 4 report the imposed modifications to the control parameters to balance the parallel currents.
Current imbalance levels (Δ
ID) are mentioned in
Table 5 and
Table 6 for both tests without and with the proposed solution. Based on the experimental results, the current curves and peak currents (
IDmax) between the parallel power devices are almost the same.
In the first test, by implementing the proposed method, turn-on and -off imbalances are reduced from 44% to 1.3% and 10.5% to 1.2%, respectively, while the static imbalance is decreased from 11% to 1.3%. In the second test, turn-on and -off imbalances are reduced from 16.8% to 1.5% and 15.9% to 1.4%, respectively, while the static imbalance is decreased from 4.2% to 0.7% (((0.05 × 100)/(3.65 + 3.6)) × 100% = 0.7%).
According to the experimental results, before applying the proposed innovative technique, a significant amount of current imbalance is shown between the drain currents during all device stages. However, the proposed method can offer a well-balanced current sharing between SiC MOSFETs by properly adjusting the correct control parameters, proving its current balancing performance against multiple impact factors and promoting the safety of the parallel SiC MOSFETs.
The pair of SiC MOSFETs that were used for the first experimental test was determined through a number of tests, connecting several pairs of SiC MOSFETs in parallel. We have reached the decision to use this particular pair of SiC MOSFETs due to the large static and dynamic current imbalances that occur. Both imbalances may be attributed to the deviation of the technical parameters between parallel devices. Such static current imbalance can only be caused due to the on-resistance difference of the SiC MOSFETs. For this reason, the elimination of the static current imbalance was only possible by driving the parallel devices with such a large VCC difference until both on-resistances of power devices became equal. In the second experimental test, the static current imbalance is much lower compared to the first case and is attributed to the asymmetrical power circuit layout. As a result, the difference between the activation voltages (VCC) of the power devices is much smaller compared to the VCC difference of the first experimental test.
Lowering VCC increases switching and conduction losses because the gate current decreases and on-resistance of SiC MOSFET increases. Therefore, it is important to suppress static current imbalance without increasing conduction losses. Static current imbalance can be minimized by equalizing the on-resistances either by increasing the on-resistance of the SiC MOSFET carrying the highest current or by reducing the on-resistance of the SiC MOSFET carrying the lowest current. However, the first static balancing approach will result in higher conduction losses compared to the second one. For this reason, in the first experimental test, the static current balancing process starts with the increase in VCC, forcing the on-resistance of the SiC MOSFET with the least drain current to become lower. However, the absolute VCC of the SiC MOSFET C2M0080120D is 25 V. As a result, the balancing process should be proceeded by decreasing the VCC of the SiC MOSFET with the highest drain current until the on-resistances of both SiC MOSFETs become equal. The VCC difference may influence dynamic current sharing, but it is compensated with the implementation of the proposed dynamic current balancing methods. As a result, the entire current imbalance is eliminated, retaining balanced switching and conduction power losses. Also, SiC MOSFETs operate under the same temperature stress level because distributed heat between power devices is kept almost equal. Since the drain-source voltage (VDS) of both parallel devices is common, due to their parallel connection, and their drain currents become equal, switching and conduction power losses are balanced.
The turn-on current ringing is attributed to the free-wheeling SiC Schottky diode due to the discharge of its self-capacitance under reverse bias. On the other hand, the turn-off current ringing is caused by the parasitic inductance of the resistive load. The drain–source voltage ringing during turn-off becomes lower with the use of the SiC Schottky diode but it cannot be dramatically reduced. Nevertheless, the switching oscillations of SiC MOSFETs could be eliminated with an RLC snubber [
44]. However, the examination of the effectiveness of the proposed method is not affected by the current ringing during the switching intervals.