**Adam Kawa \* and Robert Stala**

Department of Power Electronics and Energy Control Systems, Faculty of Electrical Engineering, Automatics, Computer Science and Biomedical Engineering, AGH University of Science and Technology, 30-059 Krakow, Poland; stala@agh.edu.pl

**\*** Correspondence: adamkawa@agh.edu.pl

Received: 2 April 2020; Accepted: 9 May 2020; Published: 13 May 2020

**Abstract:** This paper presents the research results of the bidirectional multilevel resonant switched capacitor converter (MRSCC). The converter can achieve a high voltage ratio in multilevel topology, which limits the voltage stress on switches and is able to operate with high power efficiency. The converter can be applied as an interconnector between DC voltage systems used for various applications. This paper presents a method that significantly improves the efficiency of the MSRCC through topology modification. Furthermore, the feasibility of the converter was demonstrated with the use of SiC and Si MOSFET switches, together with suitable passive components. It was demonstrated that the proposed modification of the topology makes the converter very efficient in SiC-based ones and can significantly improve the efficiency of Si MOSFET converters. The series of test results of the SiC-based converter is a novel aspect presented in this paper and shows promising achievements of efficiency. The results were obtained from the laboratory setup of 5 kW and 0.5/2 kV MRSCC. To demonstrate the bidirectional operation of the converter, a back-to-back setup (0.5/2/0.5 kV) was used. It also demonstrates that such a high-voltage gain converter can be accurately tested with the use of laboratory equipment with a typical voltage range.

**Keywords:** bidirectional converter; multilevel converter; resonant converter; SiC MOSFET; high-voltage converter; switched capacitor converter

### **1. Introduction**

The multilevel topology of the converter is favorable for high-voltage circuits due to the reduction of voltage stress on switches. Aside from typical multilevel converters, such as the flying capacitor, neutral point clamped, and cascade bridges, various concepts of switched capacitor-based converters have been proposed recently. The switched capacitor (SC) technique is suitable for high-voltage gain converters and can be effectively used in power converters, which was proven in [1,2]. On the basis of the SC technique, such multilevel topologies include MRSCC (multilevel mesonant switched capacitor converter) [3–5], modular capacitor clamped [6,7], resonant Ladder [8], MMCCC (multilevel modular capacitor clamped converter), 6X [9], converter with coupled inductors in various levels [10], and converters presented in [11–16], where the topologies include a similar concept to MRSCC.

Another important quality of a power electronic converter is its bidirectional operation capability. It is required in a vast range of applications, which incorporates battery management. The analyzed MRSCC is made up of a basic SC structure, which makes it possible to transfer energy in both directions, similar to the converters presented in [3–5,13–17].

The four-level MRSCC converter analyzed in this study (Figure 1a) is a very favorable solution when compared to well-established topologies. It comprises features of a bidirectional and multilevel topology. In MRSCC, voltage stress on the switches is equal to the voltage of a single cell; therefore, the benefit of switch voltage stress reduction is achieved. In comparison to the other concepts of SC converters, such as SCVMs (switched capacitor voltage multipliers) presented in [18,19], MRSCC is a multilevel concept that can be more favorable for high-voltage applications. Voltage stresses on some switches of SCVMs can reach the output voltage (the highest value). Due to the high voltage gain, low stress on switches, and bidirectional conversion ability, MRSCC can be a beneficial solution for the interconnection between DC voltage systems. Depending on applications, numerous DC voltage systems are often used nowadays [20], and the required DC voltage reaches 1500 V [17]. The microgrid connection is another prospective application of such a converter as well. Furthermore, in MRSCC, the energy is transferred via capacitors, which makes it possible to reduce the weight of the converter in comparison to solutions based on inductive components. In some applications, a low weight can be an important feature due to the requirements related to the assembling system components.

The basic concepts of three-level and four-level MRSCCs are presented in [3], in which a detailed analysis of losses and selection of inductance can be found. In [4], the four-level SC converter with loads connected asymmetrically was analyzed. In [5], the first harmonic approximation analysis and basic experimental results of four-level 1.2 kV MRSCC based on SiC are presented. In [13], as well as [14], the basic structure of a multilevel SC converter is used in cascaded systems. However, the problems of topology modification and efficiency improvement in ZVS conditions were not investigated.

Recently, a lot of research has focused on the elimination or limitation of *C*oss losses in different types of converters. However, very few of them present a solution for switched capacitor converters and especially multilevel ones. In [15,17], the *C*oss losses were identified as an important factor in the power efficiency limitation of the voltage doubler based on the MRSCC concept. The solution for *C*oss elimination is reported in [15] for a MOSFET converter as well as in [16] for GaN and in [17] for SiC. Therefore, the methods for achieving ZVS (zero-voltage switching) in such a converter are demonstrated in those papers, but all of them are similar and limited to two-level (voltage doubler) MRSCC. The methods rely on the application of the special control pattern to switches in higher and lower voltage levels, for additional energy delivery to the resonant inductor before dead time intervals. The energy stored in the resonant inductor is then utilized for the *C*oss voltage transition in dead time intervals. As reported in [16], this technique is problematic in light load operations, due to significant switching frequency increases. Special control with cycle skipping is required for light load operation, which itself brings some power loss and is problematic due to parasitic oscillations [21]. To mitigate this challenge, the authors in [16,17,21] proposed to move the resonant inductor to the DC side of the converter. It eliminates most disadvantages; however, it still cannot be applied to converters with a number of levels higher than 2, since in higher levels of the converter, the resonant operation could not be achieved. In this paper, the solution with an additional small inductor further referenced as the commutation supporting inductor (*L*SC) and small (commutation) capacitors connected across resonant branches is presented. This solution is free from the abovementioned limitations. It operates well at any load level, including an idle state, and can be applied for a converter with any number of voltage levels; however, it requires additional components.

*C*oss losses have an important impact on the efficiency of a high-frequency converter. A model of this type of loss in an SiC-based converter is analyzed in detail in [22]. In the case of SC converters, which contain large numbers of switches, such as SCVM [18,19], and do not operate in ZVS mode, *C*oss losses can become significant. The method for ZVS operation in SC converters with resonant inductors is an important research subject. It can be introduced to other SC topologies than MRSCC and the results of ZVS operation in an SiC-based converter presented in this paper can have even more general importance. In [23–28], SC converters with ZVS operation are presented. The presented approach for the ZVS problem solution is similar there and the converters use resonant inductors, which makes it possible to apply a phase shift switching or an operation above the resonant frequency for a charge reduction of the switch before it is turned on. The *C*oss voltage transition method and ZVS operation is accomplished in this paper for four-level MRSCC and can be extended to a higher

number of voltage levels, which is the main difference from previously presented solutions. Therefore, this paper confirms the merit of the MRSCC topology with introduced improvements. Furthermore, numerous new experimental results of Si and SiC MOSFETs based on bidirectional four-level MRSCC, such as the efficiency, voltage ratio measurements, and resonant circuit operation, are presented and compared in this paper. In essence, the main goal of this paper was to verify the concept of the topology modification towards efficiency improvement of four-level SiC-based and Si-based MRSCC converters with the major contributions of this paper being as follows:


The paper is organized as follows. In Section 2, the basics of operation of the MRSCC and simulation results are presented. In Section 3, the problem of *C*oss loss is addressed and a modification of base topology is proposed. Section 4 includes a description of the back-to-back experimental setup and experimental results of MRSCC operation.

**Figure 1.** The bidirectional four-level switched capacitor resonant DC-DC converter (**a**) the concept, (**b**,**c**) the modes of operation, (**d**) the control signals of the switches; *S*E—control signal of the even switches; *S*O—control signal of the even switches.

#### **2. Principle of Operation of the Multilevel Resonant Switched Capacitor Converter (MRSCC)**

Figure 1 presents the topology of the MRSCC. The principle of operation assumes charging and discharging of switched capacitors (*C*R) in resonant circuits and energy exchange between DC capacitors *C*1–*C*4. Each voltage level contains two transistors in HB (half bridge) configuration, with appropriate driving circuits. All the resonant branches *C*R1*L*R1-*C*R3*L*R3 are tuned to the same resonant frequency. The control of the converter, in the general concept, requires alternating the switching of odd and even switches with a 50% duty ratio with the resonant frequency of the branches *C*R1*L*R1-*C*R3*L*R3. As a result, each resonant branch (*C*Rk*L*Rk) is switched between two adjacent voltage level capacitors *C*<sup>k</sup> and *C(*k+1) (Figure 1b,c). The control signals are presented in Figure 1d.

Voltages of the capacitors *C*1-*C*<sup>4</sup> of MRSCC are nearly equalized. As a result, the voltage gain of the idealized converter in the boost-type interpretation is as follows:

$$k\_{\rm UTE} = \frac{\mathcal{U}\_{\rm P}}{\mathcal{U}\_{\rm s}} = n\_{\rm \prime} \tag{1}$$

where *U*p-higher side voltage value, *U*s-lower side voltage value, and n- number of levels.

During the *t*0-*t*<sup>1</sup> and *t*2–*t*<sup>3</sup> time intervals, the current oscillations occur in resonant branches. During the dead time intervals, the oscillations are stopped, and in the idealized and tuned circuit, the currents of resonant branches are zero during dead time. The current peak stresses of the resonant branches are given by the following relationship:

$$\begin{cases} I\_{\rm GR}(k) \text{DTmax} = \frac{\pi}{\mathcal{C}\_{\rm DT}} (n - k) I\_{\rm P} \\ I\_{\rm S} = n I\_{\rm P} \end{cases} \tag{2}$$

where *<sup>k</sup>* <sup>=</sup> 1, 2, 3, 4, ... ,(*<sup>n</sup>* <sup>−</sup> <sup>1</sup>); GDT <sup>=</sup> <sup>1</sup> <sup>−</sup> <sup>2</sup>*t*DT *<sup>T</sup>*<sup>s</sup> ≤ 1; and *I*p, *I*s-current value of lower and higher voltage side.

The current stresses of the switches are given by:

$$I\_{\rm S(2k)\rm RMS}(t) = I\_{\rm S(2k-1)\rm RMS}(t) = \frac{\pi}{2} \frac{1}{\sqrt{\rm G\_{\rm DT}}} I\_{\rm P'} \tag{3}$$

$$I\_{\rm S2RMS}(t) = I\_{\rm S1RMS}(t) = \frac{\pi}{2} \frac{1}{\sqrt{\mathbf{C}\_{\rm DT}}} (n-1) I\_{\rm P'} \tag{4}$$

where *k* = 2, 3, 4, ... , *n*, ϕ = ϕk−<sup>1</sup> = ϕk.

From Equations (3) and (4), it follows that the current stress of switches S1 and S2 is greater than others for *n* > 2. The higher the number of *n*, the bigger the difference between the current stresses of switches observed in the converter, which is an important conclusion from the switches' selection standpoint. The dead time increases the current stresses of the resonant branches and switches. It corresponds with the *G*DT coefficient variation in Equations (2)–(4). The example simulation waveforms of the 5 kW and 500 V/2 kV MRSCC are presented in Figures 2 and 3. The results were carried out from two different models and software tools. In the waveforms presented in Figure 2, during the dead time intervals, the oscillations are halted and the current of the resonant branches equals zero. In this simulation, which was performed in MATLAB/SIMULINK software, all parasitic capacitances of the switches were omitted. In a real circuit (Figure 7a), or precise behavioral simulation model designed for PSpice (Figure 3), the distortions can be observed in the dead time intervals. The distortions are caused by an interaction of resonant branches and *C*oss capacitances of the switches as can be anticipated by comparing the simulation results. Therefore, the distortions in dead time intervals are natural for this topology and are not caused by an improper design of the real converter. For PSpice simulation, the precise models of power transistors provided by the manufactures were used. The simulations parameters are consistent with the experimental setup (Tables 1 and 2) described in Section 4.


**Table 1.** Parameters for SiC and Si designs for the base configuration.

**Table 2.** The measured idle power of a single HB for different transistors used for the SiC and Si setup.


**Figure 2.** Simulation result of the 5 kW 500V/2 kV MRSCC (Multilevel Resonant Switched Capacitor Converter) in the base configuration in MATLAB/Simulink without *C*oss.

**Figure 3.** Simulation result of the 5 kW 500 V/2 kV MRSCC in the base configuration in PSpice (plotted in MATLAB) with *C*oss and other parasitics.

The transition between the boost and buck mode of operation of MRSCC does not require any special control. It depends on the relationships between voltages *U*p/*U*S. The resonant legs should be tuned to achieve the best possible performance of the converter, and only this case is analyzed in the paper. In [5], the analysis for the general case can be found. For a cost reduction, selected switches can be replaced by diodes, but only unidirectional operation is possible in this case. For example, as presented in the literature [8], a diode-based ladder resonant switched capacitor converter operates in boost mode only.

#### **3.** *C***oss Losses and Its Reduction by Application of the Commutation Supporting Inductor**

The converter operates in ZCS (zero current switch) conditions, but switching losses still occur due to charging and discharging of the output capacitances of the switches. The converter consists of a relatively large number of switches; thus, the switching losses may seriously deteriorate its efficiency, especially under the light load operation. *C*oss losses are also the subject of the research presented in [15,16], where it is eliminated by the special control of the switches. However, a solution for a converter composed of the number of voltage levels above 2 is not presented.

*C*oss losses are associated with the energy dispersion from *C*oss by the switch that is turned on as well as the power losses during *C*oss charging caused by the current flow and commutation in the corresponding switch. In an MOSFET switch, the capacitance *C*oss is strongly nonlinear and the charge transferred to *C*oss is (from the DC source *U*HB, during the lower switch turn on in a half-bridge operating with ZCS):

$$Q\_{\rm{oss}}(\mathcal{U}\_{\rm{HB}}) = \int\_0^{\mathcal{U}\_{\rm{HB}}} \mathbb{C}\_{\rm{oss\\_low}}(\mu\_{\rm{DS\\_low}}) \, d\mu\_{\rm{DS\\_low}}.\tag{5}$$

Typically, both switches in HB are the same type and have the same parameters. Therefore, the high side switch will result in the same charge transfer from a DC source. As a result, the charge *Q*oss is transferred twice in the switching period. Thus, the average value of the *C*oss losses in a half-bridge of the converter is:

$$
\Delta P\_{\rm HB\\_loss} = 2f\_{\rm s} \mathcal{U}\_{\rm HB} Q\_{\rm loss} (\mathcal{U}\_{\rm HB}) = 2f\_{\rm s} \mathcal{U}\_{\rm HB} \int\_0^{\mathcal{U}\_{\rm HB}} \mathcal{C}\_{\rm osc\\_low} \left(\mu\_{\rm DS\\_low}\right) \, d\mu\_{\rm DS\\_low} \tag{6}
$$

If *C*oss capacitances are charged to the certain voltage value in the dead time by the external circuit, the start value in the integral in Equation (6) is a nonzero value. Therefore, the losses (Equation (6)) are lower in such a case, and can even be zero (in idealized analysis). The detailed analysis of *C*oss-related energies and losses in HB can found in [29,30]. The MRSCC converter consists of *n* number of HBs composed with two various types of switches, resulting from their current stresses described by Equations (3) and (4). The total *C*oss power loss of the converter is the sum of the *C*oss losses of all HBs.

For the analysis and optimization of losses in MOSFET power transistors in MRSCC, the following FOM (figure of merit) [31] can be used:

$$F\_{\rm M} = r\_{\rm D\rm Son} Q\_{\rm oss} \tag{7}$$

where *r*DSon-turn on resistance of MOSFET, *Q*oss-charge calculated as in (5) with *U*HB = *U*S.

Based on Equations (3) and (4), the conduction loss in power transistors of the lower HB (S1, S2) and all the higher HBs (S3-S8) can be the following:

$$
\Delta P\_{\rm Rds\\_L} = I\_\text{P}^2 \frac{\pi^2}{2} r\_{\rm DSO\\_L} (n-1)^2,\tag{8}
$$

$$
\Delta P\_{\rm Rds\\_H} = I\_P^2 \frac{\pi^2}{2} r\_{\rm DSun\\_H} (n - 1) \,\tag{9}
$$

where *r*DSon\_L-on-state resistance of the S1-S2 switches, *r*DSon\_H-on-state resistance of the S3–S8 switches.

Utilizing Equations (7) and (6), the *C*oss loss in power transistors of the lower HB (S1, S2) Δ*P*Coss\_L and all the higher HBs (S3–S8) Δ*P*Coss\_H can be found:

$$
\Delta P\_{\text{Coss\\_L}} = 2f\_{\text{s}}F\_{\text{ML}} \, \text{l} \, \text{l}\_{\text{S}} \frac{1}{r\_{\text{DSun\\_L}}},\tag{10}
$$

$$
\Delta P\_{\text{Cross\\_H}} = 2f\_{\text{\textdegree}} F\_{\text{MH}} \mathcal{U}\_{\text{S}}(n-1) \frac{1}{r\_{\text{DSOon\\_H}}},\tag{11}
$$

where *rDSon*\_*L*—on resistance of the S1–S2, *rDSon*\_*H*—on resistance of the S3–S8, *fs*—switching frequency, *FML*—figure of merit (7) for S1 and S2, *FMH*—figure of merit (7) for S3–S8.

The total power loss for a given group of switches is the sum of the conduction loss and *C*oss-related loss. The selection of transistors with higher on-state resistance leads to a higher conduction loss. However, the *C*oss loss is lower in such a case, if the figure of merit (Equation (7)) is considered constant. Using the above Equations (8)–(11), it is possible to find the optimal value of *r*DSon\_L and *r*DSon\_L for which the total power loss of a transistor is minimal for a given figure of merit. Obviously, the selection of transistors is a discrete problem since only several types of devices with specific parameters are manufactured. The presented solution is continuous, but in spite of this fact, it can be used for the selection of the transistors nearest to the calculated optimal parameters. The optimal on-state resistances of transistors are given by the following relationship:

$$r\_{\text{DSon\\_L\\_opt}} = \frac{2n\mathcal{U}\_{\text{S}}\sqrt{\mathcal{U}\_{\text{N}}f\_{s}F\_{\text{ML}}}}{\pi P\_{\text{S}}(n-1)},\tag{12}$$

$$r\_{\text{DSun\\_H\\_opt}} = \frac{2n!I\_S\sqrt{\text{LI}\_N f\_s F\_{\text{MH}}}}{\pi P\_S},\tag{13}$$

where *r*DSon\_L\_opt—optimal on resistance of the S1–S2, *r*DSon\_H\_opt—optimal on resistance of the S3–S8, *f*s—switching frequency, *F*ML—figure of merit (7) for S1–S2, *F*MH—figure of merit (7) for S3–S8, *P*S—power of lower voltage side, for which value the minimum losses should occur.

The above Equations (8)–(13) were evaluated for the real setup parameters, which are presented in Section 4, especially in Tables 1 and 2. The figure of merit was calculated based on Equations (5) and (7) and the *C*oss curves from the data sheets for the two transistors selected in the initial design step: SCT3030AL as S1, S2; *F*ML = 4 nVs and C3M012090D as S3–S8; *F*MH = 5.9 nVs. The results are presented in Figure 4.

As can be observed in Figure 4, the initially selected transistors are not optimal. However, there is no significant improvement possible since it is about 20% of the total power loss in the switches. A significant improvement of efficiency may be achieved by the introduction of further modifications as described in Section 4, which nearly eliminates the total *C*oss power loss. In case of the initially selected transistors, the *C*oss loss accounts for nearly 87% of the total loss in the switches, and in the case of hypothetical optimal switches, 52%. After the introduced modification of the topology, only conduction losses are relevant. The conduction losses are lower for the initially selected transistors (SCT3030AL, C3M012090D) than in the case of any hypothetical optimal ones. Therefore, the initially

selected transistors were accepted as the final selection. After the modification, the optimization problem must be defined in another way since there is no longer a tradeoff needed between *C*oss and the on-state resistance.

**Figure 4.** Results of the optimization of losses in power switches in terms of rDSon selection.

#### *3.1. Concept of MRSCC with Commutation Inductor*

To reduce the idle power losses, the topology modification is proposed based on the application of the additional commutation inductor *L*SC, which brings a significant reduction to the switching losses. The proposed concept and commutation sequence from odd to even switches is presented in Figure 5a–c. The inductor *L*SC is connected between the output of the lowest HB and the output of the voltage divider created in the lowest voltage level. Due to the fact that all HBs are switched with a 50% duty ratio, the waveform of the current in *L*SC is symmetrically triangular with the positive and negative peaks that occur during the commutation. This current charges and discharges the output capacitances of all switches during the commutation, causing ZVS operation.

When all the HBs are turned into *DT* mode (Figure 5b), the output capacitance of all HBs is charged by the current of inductor *L*SC (HB1 directly, and the rest via *C*Ck capacitors). The transition is finished after a certain time, which depends on the peak current of the *L*SC and the values of the output capacitances of the switches. To finish the switching sequence, all the even switches are turned on (Figure 5c). Because the output capacitances of the switches are charged and discharged by an inductor, their switching losses are significantly reduced (in theory it is lossless).

**Figure 5.** (**a**–**c**) The sequence of modes for commutation from odd to even switches in MRSCC with the commutation capacitors *C*Cn and the supporting inductor *L*SC; (**d**) Simplified equivalent circuit for charging the *C*oss by *L*CS current during dead time (mode b); (**e**) further simplification.

#### *3.2. Application of Communication Capacitors in MRSCC Branches*

The resonant branches contain series inductors and therefore, for very short commutation processes, they can be considered as the current sources. The values of those current sources can be assumed equal to the instantaneous current of the resonant branches at the beginning of the commutation. In the precisely tuned circuit, those values are close to zero, which means that the resonant branches have almost no effect on the commutation process. For this reason, the commutation capacitors (relatively small) are applied to the circuit, in parallel to the resonant branches. The commutation capacitors *C*Ck can be considered as the voltage sources with the value of *U*<sup>S</sup> (as well as the all *C*<sup>k</sup> capacitors). They allow the output capacitances of higher-level HBs to be charged and discharged by the current of the *L*SC inductor flow. The inductor *L*SC can be considered as the current source with a positive (odd to even switches commutation) or negative (even to odd switches commutation) peak value. The equivalent circuit representing such conditions is shown in Figure 5d. The circuit can be further simplified, to the circuit where all *C*oss are charged in parallel and supplied by one *U*<sup>S</sup> voltage source, as shown in Figure 5e. The application of the commutation capacitors has an additional effect. In the base circuit, during the dead time, the current oscillations in the resonant branches are stopped, thus the resonant frequency of the converter is lower than the resonant frequency of the branches. The commutation capacitors clamp the resonant branches during the dead time, which allows the oscillation to continue. As a result, the resonant frequency of the converter is equal to the resonant frequency of the resonant branches. Furthermore, an additional benefit is the resonant current is a smooth sinusoidal (Figure 6b), while in the base circuit, some distortion (fast oscillation) can be observed in dead time intervals (Figures 4 and 6a). Such distortions reduce the balancing capability of the resonant branches and increase the series-equivalent output resistance of the converter, which will be presented in Section 4. The application of the *L*SC inductor and *C*Ck capacitors has no direct influence on the voltage gain of the converter. However, it improves the power efficiency and eliminates distortions of the oscillations, resulting in a reduction of the series-equivalent resistance of the converter.

Idle mode losses depend on the peak value of the current of *L*SC, which will be demonstrated in Section 4 together with the research results related to the impact of the *I*LSCpk on the switch voltage during commutation. The overall efficiency of the MRSCC is also affected by the remaining switching losses and conduction losses, which are out of the scope of this paper.

**Figure 6.** Waveforms of HB (Half Bridge) voltages and resonant branches currents for step-up steady-state operation of: (**a**) base MRSCC at 258 kHz, (**b**) modified MRSCC at 285 kHz. Experimental results of SiC-based converter 5 kW load (oscilloscope data exported and to MATLAB and plotted).
