A Comparative Study on the Switching Performance of GaN and Si Power Devices for Bipolar Complementary Modulated Converter Legs

Abstract

The commercial mature gallium nitride high electron mobility transistors (GaN HEMT) technology has drawn much attention for its great potential in industrial power electronic applications. GaN HEMT is known for low on-state resistance, high withstand voltage, and high switching frequency. This paper presents comparative experimental evaluations of GaN HEMT and conventional Si insulated gate bipolar transistors (Si IGBTs) of similar power rating. The comparative study is carried out on both the element and converter level. Firstly, on the discrete element level, the steady and dynamic characteristics of GaN HEMT are compared with Si-IGBT, including forward and reverse conducting character, and switching time. Then, the elemental switching losses are analyzed based on measured data. Finally, on a complementary buck converter level, the overall efficiency and EMI-related common-mode currents are compared. For the tested conditions, it is found that the GaN HEMT switching loss is much less than for the same power class IGBT. However, it is worth noting that special attention should be paid to reverse conduction losses in the PWM dead time (or dead band) of complementary-modulated converter legs. When migrating from IGBT to GaN, choosing a dead-time and negative gate drive voltage in conventional IGBT manner can make GaN reverse conducting losses high. It is suggested to use 0 V turn-off gate voltage and minimize the GaN dead time in order to make full use of the GaN advantages.

1. Introduction

Power electronic devices not only convert huge amounts of energy in industrial applications, but also provide the basis for upper layer control. Thus, efficiency and switching frequency are important factors [1,2]. Compared with silicon (Si)-based power devices, such as conventional MOSFETs and IGBTs, gallium nitride high electron mobility transistors (GaN HEMTs) have gained attention recently for combining the different advantages of MOSFET and IGBT, offering great potential of enabling higher switching frequency, higher efficiency, and higher power density [3,4,5,6,7]. Conventionally, MOSFETs and IGBTs differ in three main aspects: Firstly, the major difference is in conducting character. The MOSFET conducting character is resistor-like, so the conduction loss increases as the square with respect to the current. In such a case the maximum drain current is limited by power dissipation caused by R_ds(on). The IGBT conducting character is diode-like, the voltage drop is almost constant and the loss is proportional to the current, so the IGBT can conduct much higher current with limited power losses. Secondly, MOSFETs are normally used for low-voltage applications. To withstand high voltages, R_ds is much larger and the power dissipation caused by R_ds limits the possible current increase. Although there are MOSFETs rated at 600 V and 1000 V, the current is usually less than in a GaN HEMT, and the R_ds is at least two times or more than that of a GaN HEMT. Thirdly, the IGBT suffers from long-duration tail current during switching off, which limits the use of IGBTs in high switching frequency applications. To sum up, GaN can offer the advantages of both MOSFETs and IGBTs for high-frequency, low-loss, high-voltage applications.

In the literature, implementations of GaN devices in various applications are reported [8,9], such as motor drives, DC/DC converters, battery chargers, etc. Various studies were carried out on GaN HEMTs from different perspectives. The different material compositions and manufacturing processes of GaN HEMT are studied in [10,11]. In [12,13], the calculations for GaN HEMT losses, including switching losses, parasitic capacitance losses, etc., are presented. The reliability assessment of GaN HEMTs for power switching applications is given in [14,15]. In [14] an extensive analysis is provided on the physical mechanisms responsible for the failure of GaN HEMTs, and several critical characteristics are recommended to evaluate the electrical reliability of GaN HEMTs. In [15], the most relevant mechanisms that limit the dynamic performance and the reliability of GaN HEMTs are discussed.

However, as a main difference from the conventional Si-based devices, the reverse conducting character of GaN HEMT does not require an additional diode and the reverse conduction voltage drop is largely related with the gate-drive voltage [16,17].

This reverse conduction is seldom discussed in detail in the literature. In [18], GaN device loss including reverse conducting loss is compared with different bipolar and unipolar modulation strategies for a H-bridge power converter driving a DC motor. However, in [18], the reverse conduction is mainly evaluated for always-on or always-off switching devices, and the loss induced by the dead-band in a complementary modulated converter leg is neglected.

Actually, for complementary-modulated converter legs, the dead band must be set to avoid DC bus short circuits. In this dead zone, if not properly configured, the GaN transistor voltage drop can be relatively large for dead band reverse conduction, inducing higher losses during this period.

In this paper, a comparative study is carried out on similar rating GaN HEMTs and IGBTs. The comparative study is carried out on both the element and converter level. Firstly, on a discrete element level, the steady and dynamic characters of GaN HEMTs are compared with those of Si-IGBTs, including forward and reverse conducting character, and switching time. Then, the elementary switching losses are analyzed based on the measured data. Finally, on the buck converter level, the overall efficiency and EMI-related common-mode current performance [19,20] is compared.

The rest of the paper is organized as follows: Section 2 describes the experimental platform. Section 3 provides a comparative study on the steady and dynamic characteristics of Si-IGBTs and GaN HEMTs. Section 4 analyzes device losses. Section 5 gives the overall efficiency and EMI performance on a buck converter level. Finally, Section 6 concludes this paper.

2. Experiment Platform

In this paper, 600 V rating devices are compared, including GaN HEMTs and IGBTs at similar power ratings. The parameters of the two devices are listed in

Table 1. Parameters of the two devices.

Device	Rated Voltage	Rated Current	Conduction Characteristic	Reverse Conducting Voltage Drop (VF)
GS66516T	650 V	60 A	RDS(on) = 25 mΩ	5.28 V ($V_{GS}$ = −3 V) 0.2 V ($V_{GS}$ = 6 V)
FS50R06W1E3	600 V	50 A	$V_{CE}$ = 1.60 V	1.50 V

.

The GaN device is a GS66516T and the IGBT is a FS50R06W1E3. The rated current differs by about 10%, which was due to the device availability as it is difficult to find two devices with identical ratings. Device pictures are shown in Figure 1.

The experimental rig is shown in Figure 2. The inverter leg is driven by corresponding drive circuits with PWM signals generated by a controller. The load is a power variable resistance box, which is able to adjust the output current. The inverter leg is configured as a full bridge complementary modulated buck converter. The gate drive resistance is chosen as 10 Ω. The oscilloscope is a DLM2054 (Yokogawa, Tokyo, Japan) with 500 MHz bandwidth. The current probe is a Yokogawa 701932 with 100 MHz bandwidth. The voltage probe is a P5200A high-voltage differential probe (Tektronix, Beaverton, OR, USA) with 50 MHz bandwidth. Even compared with a 500 MHz probe, the used 50 MHz probe can accurately measure a 14 ns step signal rising time within 1% error. According to the experimental tests, the used voltage probe has 14 ns propagation delay, which is similar as that of 13 ns for the current probe. The similarity in propagation delay can help reduce loss calculation errors. Before each test, the current probe is demagnetized and the offset is adjusted to be less than 1 mA.

3. Comparative Study of Elementary Static and Dynamic Characteristics

This section presents an experimental study on the characteristics of the GaN HEMT and Si-IGBT, including forward and reverse conducting character, and switching time.

3.1. Static Characteristic

The GaN HEMT static characteristics are similar to those of a traditional Si MOSFET device at a higher voltage level, but the reverse characteristics are quite different. There is no diode connected in parallel with the power switch. The reverse freewheeling current can be conducted by GaN transistor itself. Its reverse characteristic is similar to that of a diode, but it is greatly affected by the gate-to-source voltage ($V_{GS}$).

Figure 3 gives the experimentally measured reverse conducting voltage drop of the GaN transistor with different $V_{GS}$ of 6 V, 0 V and −3 V, respectively. It can be seen that the reverse characteristic of the GaN HEMT is greatly affected by the gate voltage. When the gate voltage is 6 V, its reverse characteristic is almost a pure resistance characteristic. The resistance value is about 33 mΩ. When the $V_{GS}$ is 0 V or negative, its characteristic is similar to a diode (or a diode plus resistor). From the slew rate of the curve, it can be seen that the resistance increases as the gate voltage decreases. The smaller the $V_{GS}$, the higher the reverse conducting loss. The reverse conducting voltage is related with both $V_{GS}$ and conducting resistance. $V_{GS}$ plays a more important role than the conducting resistance.

In order to explore $V_{GS}$ influence on the freewheeling, device Q2 is fully measured under different $V_{GS}$ conditions. The experimental waveforms are shown in Figure 4. It can be seen that under the same bus voltage, the $V_{GS}$ influence is very obvious. When $V_{GS}$ = 6 V, the reverse conducting voltage drop of GaN HEMT is 0.04 V. When $V_{GS}$ = −3 V, the reverse conducting voltage drop increases to 4.65 V, signifying a sharply increased reverse conducting loss.

In practice, both 0 V and −3 V $V_{GS}$ for turn-off exist for different purposes. V_GS = 0 V can provide a lower reverse loss, while −3 V is for more robust gate drivea and better noise immunity to avoid turn-ons by mistake.

3.2. Dynamic Characteristics

The GaN HEMT switching transient is fast. The dynamic performance is experimentally tested with a fixed bus voltage. The transient time (turn-on, turn off) is defined as the settling time to reach 90% of the steady state value (in other words, after this instant, the waveform remains in a 10% error range of the steady state value). The GaN HEMT and IGBT turn-on waveforms are given in Figure 5. It can be seen that GaN measurements suffer from more severe oscillations, which could be due to two reasons. Firstly, the GaN packaging is characterized by a low parasitic inductance, and this smaller “electrical inertia” can more easily oscillate under stimulation. Secondly, the GaN switching transient is faster, inducing stronger stimulation for oscillations.

By varying the load resistance, the relationship between turn-on time and current can be obtained and is shown in Figure 6. For GaN housing/packaging, as the parasitic inductance is quite small and the switching transient is very fast, severe oscillations can be induced in the parasitic parameter and this affects the measurement of real switching time. To resolve the oscillation problem, a low pass filter provided by the oscilloscope function with 4 MHz bandwidth is applied to filter I_DS for time measurements, as indicated in the figure.

It can be seen that as the current increases, the IGBT turn-on time increases significantly, while that of the GaN HEMT remains almost unchanged. At the same current, the turn-on time of the GaN HEMT is almost four times or much faster than that of the IGBT, which shows that GaN HEMT has high-speed turn-on capability.

Figure 7 shows GaN and IGBT turn-off waveforms. Figure 8 shows the relationship between the turn-off speed and the bus current. It can be seen from the Figure 8 that as the current increases, the turn-off time is significantly reduced. This is an interesting phenomenon that is often ignored. It is especially obvious for the turn off phase. Due to parasitic capacitance between the gate-source (or gate-emitter) and gate-drain (or gate-collector), the device exhibits an output capacitance on the load. When the load current increases, the output capacitor discharges more quickly during the turn off transient. Through the equivalent capacitive connection, the gate voltage drops more quickly, and the turn off time is shortened.

Like the turn-on case, the turn-off transient of the GaN HEMT is much faster than that of the IGBT, which shows that the GaN HEMT has high-speed turn-off capability.

4. Element Loss Measurement and Analysis

This section gives the detailed loss measurement and comparison of GaN HEMT and IGBT power switching elements. The compared losses include switching losses, reverse conduction losses and forward conduction losses.

4.1. Switching Losses

The switching loss of a power device is a major part of the device loss, and it is an important index of the power device. The switching loss formula during turn-on (or turn-off) is:

$$E_{on(off)}={\displaystyle {\int }_0^t{V}_{DS}{I}_{DS}dt}$$

(1)

where t indicates the switching transient time. For IGBT, the corresponding voltage and current should be $V_{CE}$ and $I_C$.

The corresponding voltage and current are recorded by the high bandwidth oscilloscope and probe. The switching loss is calculated by the Euler discrete integration method, as in (2):

$$E_{on}={\displaystyle {\int }_0^t{V}_{DS}{I}_{DS}dt}={\displaystyle \sum _{i=1}^n{V}_i{I}_{i}T}$$

(2)

where T is the sampling time interval and T = 0.8 ns.

Figure 9 shows GaN HEMT and IGBT turn-on waveforms, respectively. The oscillation is due to parasitic measurement parameters and does not affect the integral loss calculation, since the average value of the high frequency oscillation is almost zero after integration. The multiplier item $V_{DS}$ $I_{DS}$ for GaN HEMT is also calculated by the oscilloscope, which is used to calculate the loss. The relationship between turn-on loss and current is obtained by changing the load current, as in Figure 10. It can be seen that the IGBT turn-on loss increases more rapidly as the current increases, and the GaN HEMT turn-on loss only increases slightly.

Figure 11 shows GaN HEMT and IGBT turn-off waveforms respectively. Turn-off loss is obtained in the same way as in Figure 12. The experimental data shows that the turn-off loss of the GaN device is also much lower than that of the IGBT.

As can be seen from Figure 13, the switching loss is proportional to the switching frequency f. At the same frequency, the switching loss of the IGBT is almost 10 times the loss in the GaN HEMT. In other words, given the same switching loss, the GaN HEMT switching frequency can be 10 times that of IGBTs, which is a significant advantage for improving the control bandwidth.

4.2. Reverse Conduction Loss

When the GaN HEMT is working in reverse conducting (freewheeling) mode, $V_{GS}$ should be supplied with a forward voltage to reduce the loss. However, in the full bridge topology, a dead band must be set to avoid DC bus short-circuits, as shown in Figure 4a. The dead time is not only dependent on the power-switching element, but also is related with the external circuit. Both the inductance value and loading conditions have an influence on the dead time selection. A 2 µs dead time is selected for the IGBT according our experimental platform configuration with enough security margin to avoid any damage. During the dead time, the $V_{GS}$ is negative, the voltage drop across the GaN HEMT is large, and the loss during this period is larger. Different GaN dead times and gate voltages are used in the comparison.

Since the voltage drop across the GaN HEMT is different during the dead zone, the reverse loss calculation formula is:

$$P_{GaN}=(2(V_{TH}+R_{DSon}{I}_{DS})I_{DS}{T}_D+R_{DSon}{I}_{DS}{}^2{T}_{ON})f$$

(3)

where $V_{TH}$ is 1.2 V ($V_{GS}=0 \mathrm{V}$) or 4.2 V ($V_{GS}=-3 \mathrm{V}$) for the tested GaN HEMT, $T_{ON}$ is the turn-on time, $T_D$ is the dead-band time, $f$ is the PWM frequency. The coefficient 2 refers to the fact that there are two dead-time periods in one PWM cycle for this experimental configuration, as Q2 always conducts the freewheeling current before Q1 or Q2 turn on.

IGBT freewheeling loss is:

$$P_{IGBT}=V_{F}I(T_{ON}+2T_D)f+2\times \frac{1}{2}{Q}_{rr}{U}_{DC}f$$

(4)

where $V_F$ is the reverse conduction voltage, $I$ is the freewheeling current, $Q_{rr}$ is the diode reverse recovery charge, $U_{DC}$ is the DC bus voltage. The relationship between reverse loss and f is compared under rated conditions. The rated condition parameters are listed in

Table 2. Rated conditions.

Parameters	${\mathit{U}}_{\mathit{D}\mathit{C}}/\mathbf{V}$	IDC/A	${\mathit{T}}_{\mathit{D}}/\mathsf{\mu }\mathbf{s}$
Value	300	50	2 or 0.2

. The loss results are given in Figure 14.

It can be seen from Figure 14a that if GaN dead time is chosen as a conventional IGBT, the reverse loss could be similar to that of the IGBT. This might be a possible choice when migrating from IGBT to GaN. Since GaN can switch much faster than IGBT, reducing the GaN dead-time and using a 0 V turn-off gate voltage can keep the GaN loss at a low level and make full use of the GaN advantages, as shown in Figure 14b. Thus, when migrating from IGBT to GaN, more attention should be paid to the dead time and drive voltage.

4.3. Forward Conduction Loss

The GaN conduction resistance is very small, which makes its conduction losses comparable to IGBTs under high current conditions. Due to the device characteristics, the conduction loss of IGBTs is proportional to the current flowing through them, while the conduction loss of GaN is proportional to the square of the current, as seen in (5):

$$\begin{array}{l}{P}_{on(GaN)}=I^2{R}_{DS(ON)}D\\ P_{on(IGBT)}=V_{CE}ID\end{array}$$

(5)

where D is the duty value of the switch. In this experiment, D = 0.5.

Figure 15 shows the comparison of the losses of both devices. It can be seen from the figure that the conduction loss of GaN is smaller than that of the IGBT in most ranges. However, for larger currents, the advantage decreases.

5. Converter Performance Analysis

After analyzing the element performance in the above sections, the performance of a complementary modulated buck converter leg is comparatively evaluated in this section, including overall efficiency and common mode noise.

5.1. Overall Efficiency of the Complementary Buck Converter Leg

As aforementioned, the buck converter leg topology is shown in Figure 2. The buck converter leg using GaN HEMT and Si IGBT was tested under the same loading conditions. The efficiency result is shown in Figure 16. It can be seen the efficiency of the GaN HEMT buck converter is higher than that of the IGBT one, by an average of two percentage points in the tested conditions.

5.2. Common Mode Interference

The high voltage and current changing rate (dv/dt and di/dt) in turn-on and turn-off transients are the main source of high-frequency electromagnetic interference (EMI). The generated EMI noise mainly affects the normal operation of the peripheral electronic devices through common mode coupling. The GaN HEMT has high-speed switching capabilities, resulting in high dv/dt and di/dt, so the EMI is also evaluated through experiments. The test configuration is shown in Figure 17.

Experimental tests measure the common-mode current for both the GaN-HEMT and Si-IGBT systems under the same conditions. The turn-on waveforms were recorded as shown in Figure 18. Comparing the two devices, it can be found that the gate-drive signal of the GaN HEMT rises to the rated value more quickly. This is because of its smaller gate capacitance. The rising speed of the IGBT drive signal is relatively slower. For the two devices, both common-mode current amplitude and duration are almost at the same level.

Figure 19 shows the gate-drive signal and common-mode current for both devices during turn-off. Similarly, the common-mode current amplitude of the GaN HEMT is almost the same as that of the IGBT, but with higher frequency oscillation. To sum-up, the GaN HEMT and Si IGBT devices have almost the same level of common mode noise during turn-on and turn-off transients, although the GaN HEMT can switch much faster.

6. Conclusions

This paper presents a comparative study of same rating GaN HEMTs and IGBTs. The characteristics, losses, and performance in converter efficiency and EMI are experimentally evaluated using a complementary-modulated converter leg. Generally, the GaN HEMT switching loss is much less than that of the IGBT. It is also found that the GaN HEMT common mode current level is similar to the IGBT one, although GaN switches much faster.

For a complementary-modulated converter leg, a dead band is usually set to avoid DC bus short circuits. When migrating from IGBT to GaN, choosing a dead-time and negative gate drive voltage in a conventional IGBT manner can make GaN reverse conducting loss high. It is suggested to use a 0 V turn-off gate voltage and minimize the GaN dead time in order to reduce the losses and make full use of the GaN advantages.

To sum-up, GaN HEMTs have great potential for high-speed, high-power density applications. In complementary converter-legs, attention should be paid to minimizing the reverse conduction loss.