SiC/GaN power semiconductor devices: a theoretical comparison and experimental evaluation under different switching conditions

(This study is for special section ‘Design, modelling and control of electric drives for transportation applications’) The conduction and switching losses of silicon carbide (SIC) and gallium nitride (GaN) power transistors are compared in this study. Voltage rating of commercial GaN power transistors is <650 V, whereas that of SiC power transistors is <1200 V. This study begins with a theoretical analysis that examines how the characteristics of a 1200 V SiC metal–oxide–semiconductor field-effect transistor (MOSFET) change if device design is re-optimised for 600 V blocking voltage. Afterwards, a range of commercial devices [1200 V SIC junction gate FET, 1200 V SiCMOSFET, 650 V SiC-MOSFET and 650 V GaN high-electron-mobility transistor (HEMT)] with the same current rating are characterised and their conduction losses, inter-electrode capacitances and switching energy Esw are compared, where it is shown that GaN-HEMT has smaller conduction and switching losses than SiC devices. Finally, a zero-voltage switching circuit is used to evaluate all the devices, where device only produces turn-OFF switching losses and it is shown that GaN-HEMT has less switching losses than SiC device in this soft switching mode. It is also shown in this study that 1200 V SiC-MOSFET has smaller conduction and switching losses than 650 V SiC-MOSFET.


Introduction
Electrical vehicle (EV) is an essential technology in the global fight to reduce environmental pollution and harmful gas emissions [1].Power electronics systems are important for electrical energy conversion within EVs [2], where power semiconductor devices play an important role.Understanding power semiconductor devices characteristics are crucial for engineers to design high efficiency, high power density power converters so as to improve overall performance of EVs such as increase range and reliability.
Wide bandgap power semiconductor devices such as silicon carbide (SiC) and gallium nitride (GaN) have recently become a hot research topic because they are able to operate in higher temperature, higher frequency and realise higher energy conversion efficiency in comparison with traditional silicon (Si) power semiconductor devices.Commercial SiC transistors (JFET, MOSFET) can block voltage above 1200 V and GaN transistor (HEMT) is able to withstand a maximal voltage of 650 V, while they can conduct current from a few amperes to a few tens of amperes.Both SiC and GaN devices can be applied in EVs, in which the voltage rating of different electrical systems is found to be from low voltage to high voltage.Low voltage is normally from 12 to 42 V and mainly for vehicle electrical equipment, where GaN devices can be used; while high voltage can reach up to 600 V and is mainly for vehicle motor drive, where SiC devices can be used because of their high voltage rating.The voltage rating of battery pack in an EV can vary from 200 to 400 V, and it is found in the literature that both SiC and GaN devices are applied in vehicle based battery charger [3,4].
When power semiconductor devices convert energy, they produce losses.Knowing SiC and GaN power devices losses is helpful in order to choose appropriate devices for vehicle-based power electronics systems.It is reported in the literature the experimental comparisons between SiC and Si power semiconductor devices [5] or between GaN and Si power semiconductor devices [6].However, few publications are focused on experimental comparison between SiC and GaN power semiconductor devices due to their voltage rating mismatch.
Previous study on device losses comparison has been reported by the authors in [7,8].The objective of this paper is to theoretically analyse SiC devices losses change when blocking voltage reduces from 1200 to 600 V and then experimentally evaluate commercial SiC and GaN devices switching losses in different switching conditions.Using soft switching technique such as zero-voltage switching (ZVS) helps to reduce device switching losses, so all the devices are also compared under soft-switching condition, which presents more experimental results than previous work in [7,8].
The paper is structured as follows: at first, SiC power devices conduction loss and switching loss change when reducing device blocking voltage is theoretically analysed.Afterwards, different commercial SiC and GaN power devices ON-state resistance and inter-electrode capacitances values are compared.Then, those devices switching energy under hard and soft switching conditions are compared and conclusions are given finally.

Conduction loss comparison
As shown in Fig. 1a, ON-state resistance R ON of a MOSFET is mainly composed by device channel resistance R ch and drift region resistance R drift .
This relationship can be expressed in terms of specific resistance (m Ω ⋅ mm 2 ) obtained by multiplying resistance by device active area in the following equation: Demonstrated in [9], device minimal R drift, sp is proportional to device maximal blocking voltage V DSS 2.5 .Meanwhile, device R drift, sp is also proportional to device drift region thickness W D .Thus, device R drift, sp and W D values of a 600 and a 1200 V devices follow the same relation, which is given as As shown in [10,11], device specific channel resistance R ch, sp is reversely proportional to channel mobility μ ch and it is almost independent on V DSS voltage, so device R ch, sp does not change when V DSS voltage reduces.
For a 1200 V SiC-MOSFET, due to the relatively low electron mobility value of SiC material [Electron mobility of SiC varies from 400 to 900 cm 2 /Vs depending on SiC polytypes, which is smaller than Si (around 1300 cm 2 /Vs).],R ch, sp is about 40% of the total R ON, sp if all the parameters are applied the values given in [12].Thus, R ON, sp value of a 600 V device is about a half that of a 1200 V device It is shown in Fig. 1b one commercial SiC-JFET structure from Infineon, which is quite similar to MOSFET.There are both lateral and vertical channels inside this JFET, so its R ON, sp are constituted by R drift, sp , specific lateral and vertical channel resistances (R ch, L and R ch, V ).If all the parameters are applied the values givens in [13], R ON, sp of a 600 V device is about 60% of a 1200 V device.
For another type of SiC-JFET (from Siemens) without lateral channel, it is found that R ON, sp of a 600 V device is half that of a 1200 V device if using all the parameters given in [14].
It can be summarised that when device blocking voltage reduces from 1200 to 600 V, device R ON, sp also reduces approximately half.

Switching loss comparison
Device inter-electrode capacitances C gd , C ds and C gs between each terminal are illustrated in Fig. 1.Unlike C gs , C gd and C ds are V DS voltage dependent capacitances and their values can be approximately calculated by the following equation: where C x refers to either C gd or C ds , A x refers to each capacitance active area and W S is depletion region thickness, which is dependent on device switching voltage V S .Active area A x can be obtained by multiplying a device- dependent constant, b, to the device area A. As given in [9], following equation can be used to show the relation between W D and W S : By combining above (4) and ( 5), each capacitor stored charge Q x during switching is obtained If 600 and 1200 V devices switch at the same voltage, their specific charge (Q x, sp ) comparison can be obtained by following equation: By combining (2) and ( 7) together, it is shown in the following equation that unlike device R ON, sp , Q x, sp of 600 V device is four times bigger than 1200 V device It is shown in Fig. 2 power transistor ideal switching waveforms when device switches at voltage V S and current I S .The transistor gate-drain charge Q gd plays an important role in device switching, because its discharge and charge time t by gate current I g determines switching speed which directly influences device switching losses E sw .The following equation can be used to approximately calculate device E sw of one switching period by supposing that device has the same turn-ON and turn-OFF switching losses where V com is the controlled gate voltage, V pl is the Miller-plateau voltage and R g is the gate resistor.It is to be noted that device output capacitance C oss (which is the sum of C gd and C ds ) stored energy E oss is dissipated in device channel when it is switched ON and E oss is recovered when it is switched OFF.By adding E oss in calculated turn-ON switching loss and subtracting it from calculated turn-OFF switching loss, device total switching loss can still be estimated by (9).
Device E sw is shown to be proportional to its Q gd .By normalising for device active area, 600 and 1200 V device specific switching loss E sw, sp shall follow the same relation as their specific charge Equation ( 10) can be used to compare switching losses between 600 and 1200 V devices of the same current rating only when the relative difference in active area is known.This is considered in the following subsection.

600 V/1200 V Device comparison
Device conduction current I D capability is determined by device maximal junction temperature T j(max) , which can be obtained by the following equation: Initially, only the thermal resistance R th of the die is considered (without device packaging influence), which is given by the following equation showing that it is determined by device thickness (which is assumed to be equal to device drift region thickness W D ), active area A and material thermal conductivity k.Thus By combining (11) and ( 12), I D can be obtained If 600 and 1200 V devices have the same current rating I D , k and T j(max) , their comparison on device area A is expressed by the following equation: By multiplying ( 14) to ( 10), 1200 and 600 V devices switching losses can then be obtained in the form of the equation below: In another condition where device packaging influence is considered, device R th is mainly determined by packaging type, where its area and thickness have little contribution on R th value [15].In this condition, as 600 and 1200 V devices have the same R th value, their R ON should be identical if their current rating is the same, so A 600 V is half that of the A 1200 V .Their switching loss comparison is then shown by the equation below: It is found that 600 V SiC device switching loss is bigger than 1200 V device in both conditions if their current rating is the same.
In the following section, in order to validate the theoretical analysis, SiC and GaN power devices will be experimentally evaluated in order to compare their losses.

Device characteristics comparison and measurement
ON-state resistance R ON datasheet values of a range of commercial device (listed in Table 1) with similar 30 A current rating are compared in Fig. 3a, which shows that R ON of GaN-HEMT is 25% lower than that of SiC device.R ON of 650 V SiC-MOSFET is even 50% higher than 1200 V device.
The theoretical study in Section 2.1 shows that in comparison with a 1200 V SiC device with the same current rating, R ON of a 600 V SiC is bigger than 1200 V device if device active area reduces more than a half [see (3)].Another 600 V/40 A SiC-MOSFET device [16] (GP1T072A060B, 600 V/40 A, V th ≃ 2.8 V) shows that its R ON is about 72 mΩ, which is still bigger than GaN device of the same current rating.Device R ON values can be obtained in the experiments either from a curve tracer or from an electrical circuit (shown in Fig. 3b), where device voltage V DS and conduction current I D are measured when they stabilise after applying current pulse I.
Datasheet values of the inter-electrode capacitance comparison of the above devices are shown in Fig. 4. It is found that device input capacitance C iss = C gs + C gd of 650 V SiC-MOSFET is slightly bigger than 1200 V SiC-MOSFET, and they are bigger than that of GaN-HEMT.Reverse transfer capacitance C rss = C gd of GaN-HEMT is much smaller than that of SiC devices and it is also shown that 650 V SiC-MOSFET C rss is bigger than 1200 V SiC-MOSFET when bias voltage is beyond 20 V. Device output capacitance C oss = C ds + C gd values of the aforementioned devices are similar, among which GaN-HEMT still has the smallest C oss value when V DS is superior to 100 V.
If one compares the datasheet of the above 600 V/40 A SiC-MOSFET with a 1200 V/32 A SiC-MOSFET [17] (GP1T080A120B, Vth ≃ 2.8 V) of another manufacturer (Global Power Technologies Group), it can be noted that inter-electrode capacitances of the former device are also bigger than the latter one.
The inter-electrode capacitances values of the GaN-HEMT in Fig. 4 are given on a linear-linear scale in the device datasheet, where it is difficult to extract C rss value due to the strong non- linearity of its value exceeding more than two orders of magnitude.Power semiconductor device inter-electrode capacitances can be measured by small signal method, where device is biased by external DC voltage and small signal can be either generated by an impedance analyser [18] or by a vector network analyser [19].
One solution to measure device inter-electrode capacitance is shown in Fig. 5, where an impedance analyser (IA) and ground connection is used.Devices V DS and V GS are biased by external voltage sources.Three external capacitors are used to block the DC voltage between device terminals with IA connector and ground, while their impedance is neglected when passing high frequency (MHz range) AC current.High impedance branches are constituted by three big resistances, which guarantee that all the AC current flow through the transistor.In the electrical circuit shown in Fig. 5, only current flowing through C gd is measured by IA, because the current flowing through capacitance C ds flows to the ground.By varying V DS voltage, device C gd values of different biased voltages are measured.Other measurement circuits can be used to measure C gs and C ds capacitances with similar measurement principle.
By knowing device inter-electrode capacitances, device switching losses can be measured and results will be presented in the following subsection.
The electrical circuit of the switching mesh is shown in Fig. 6a, in which it is constituted by a bus capacitor C bus , a half bridge circuit with two power semiconductor devices S1 and S2 together with their drivers.Gate mesh and switching mesh of each device are minimised in order to reduce gate mesh inductance L para, g and switching mesh inductance L para, sw .Lower device drain switching current I D and drain source switching voltage V DS are measured to calculate the device switching energy.
The realisation switching circuit to test 1200 V SiC-JFET is shown in Fig. 6b, in which the die of the device is mounted in a copper substrate.The device is switched with a gate voltage from −18 to 0 V. L para, g can be obtained by measuring gate voltage transient waveform in order to get resonance frequency, which is supposed to be due to resonance between L para, g and device input capacitance C iss .L para, g in the SiC-JFET switching circuit is about 2nH.An AC current probe (P6022, 1 kHz-120 MHz) is used to measure I D and an active differential voltage probe (TA043, 100 MHz) is used to measure V DS .
The realisation switching circuit of 1200 and 650 V SiC-MOSFET is shown in Fig. 6c, where the switching circuit is the same except the packaging type of 1200 V device is TO-247 and that of 650 V device is TO-220.The same gate voltage from −5 to 20 V is used to drive both devices and L para, g in the switching circuit is also about 2nH.
Finally, the realisation switching circuit of 650 V GaN-HEMT is shown in Fig. 6d, where device is in GaNPX package.Gate voltage from 0 to 7 V is used to drive the device and measured L para, g in the prototype is around 3nH.
The lumped gate resistance in all the measurements is all around 4 Ω, which is composed by gate driver output resistance, external gate mesh resistance and device internal gate resistance.For SiC-MOSFET and GaN-HEMT, devices are driven by the same gate driver IXDN609SI and I D is measured by a 1.2 GHz current shunt (SSDN-414-025) while V DS is measured by the same differential voltage probe.A high bandwidth oscilloscope up to 1.5 GHz is used in the measurement.Switching energy of all the devices is measured by double pulse test in hard switching conditions at room temperature.

Device switching energy comparison in hard switching:
The switching waveforms comparison when devices switch at V DS = 200 V and I D = 5 A are shown in Fig. 7, whereas  The turn-ON transition for GaN-HEMT is indicated from t 1 to t 3 in Figs.7a and 8a.During t 1 -t 2 , when current I D starts to rise, the presence of L para, sw induces a voltage drop L para, sw ⋅ (dI D /dt) of V DS voltage.During t 2 -t 3 , V DS voltage decreases to device ON-state voltage V DS(on) .Similar turn-ON transition of other devices can be observed in their waveforms.
The turn-OFF transition for GaN-HEMT is indicated from t 6 -t 8 in Figs.7b and 8b.During t 6 -t 7 , when S1 voltage V DS starts to rise, S2 C oss starts to discharge, which reduces S1 I D current value, because load current is constant.During t 7 -t 8 , S1 I D starts to decrease to zero, the presence of L para, sw induces a voltage overshoot L para, sw ⋅ (dI D /dt) of V DS voltage.Similar turn-OFF transition of other devices can be observed in their waveforms.
As shown in the results, when V DS = 200 V, V DS turn-OFF transition time (from 10 to 90% switching voltage) of 1200 V SiC-MOSFET is around 12 ns while that of 650 V SiC-MOSFET is around 15 ns.When V DS = 300 V, V DS turn-OFF transition time of 1200 V SiC-MOSFET is around 15 ns while that of 650 V SiC-MOSFET is around 16 ns, indicating that 1200 V SiC-MOSFET switches faster than 650 V SiC-MOSFET in turn-OFF switching, which is supposed to be because the transfer capacitance C rss of 1200 V device is smaller than 650 V device in high voltage.
In terms of turn-ON switching, when I D = 5 and I D = 10 A, current transition time (from 10 to 90% peak current) for both 1200 and 650 V SiC-MOSFET is quite similar, which is around 7 and 8 ns separately, suggesting a similar device input capacitance C iss , which corresponds with their C iss datasheet value (C iss of 1200 V device is slightly smaller than 650 V device).
When comparing with 650 V GaN-HEMT, it is shown that GaN-HEMT current turn-ON transition time is around 5 ns in both switching conditions, which confirms the datasheet value that C iss of GaN-HEMT is only about one-sixth of 1200 V SiC-MOSFET.In terms of turn-OFF switching, when V DS = 200 and V DS = 300 V, GaN-HEMT voltage transition time is around 8 and 6 ns separately, confirming its smaller C rss value than SiC devices.
When comparing SiC-JFET with GaN-HEMT, it is observed that both V DS and I D transition time of GaN-HEMT is shorter than SiC-JFET in the above switching conditions, which also confirms that GaN-HEMT has smaller inter-electrode capacitances than SiC-JFET.It can be thus concluded that GaN-HEMT switches faster than SiC power transistors and 1200 V SiC-MOSFET switches faster than 650 V SiC-MOSFET.
It is shown in Fig. 9 the switching energy E sw comparison results of all the aforementioned devices when they switch at both 200 and 300 V.It is to be noted that device C oss discharge current is excluded in the measured I D current when device is switched ON, so device C oss stored energy E oss is dissipated and it is excluded in obtained turn-ON switching energy.In contrary, C oss charge current is included in I D current when device is switched OFF, so E oss is included in obtained turn-OFF switching energy.
It is shown in those results that E sw of 1200 V SiC-MOSFET is smaller than that of 650 V SiC-MOSFET, which confirms the theoretical analysis.When compared to SiC device, E sw of GaN- HEMT is even smaller, which shows that it is more suitable for below 300 V electrical energy conversion than SiC devices.
It is also shown that obtained GaN-HEMT turn-ON E sw is very close to SiC-MOSFET at some switching conditions.This is supposed to be the difference of L para, sw in device switching mesh, which is due to the difference of device packaging type.L para, sw value in SiC-MOSFET switching mesh is measured to be 80 nH, which is bigger than measured 36 nH in GaN-HEMT switching mesh.Bigger L para, sw value causes bigger voltage drop across it when I D is rising at turn-ON switching, which decreases overlapping time between switching voltage and current.This snubber effect results in a small measured turn-ON switching energy value in SiC-MOSFET switching circuit.
For both GaN-HEMT and 1200 V SiC-MOSFET, device turn-ON E sw are larger than turn-OFF E sw and therefore those devices will benefit from ZVS where turn-ON losses can be eliminated.Further switching measurements for these devices under ZVS condition will be presented in the following section.In power electronics systems, normally-OFF devices are more preferable than normally-ON devices because of safe reason in case of gate driver fault, so normally-ON SiC-JFET is not considered in the following analysis.

Device switching energy comparison in soft switching:
It is illustrated in Fig. 10a device soft switching circuit.Unlike device hard switching circuit, another external power supply is used to maintain output capacitor C out voltage constant.Devices S1 and S2 areas are controlled by the signal as shown in Fig. 10b.
At instant t1, device S2 is switched ON and device S1 remains OFF, so output inductor L is started to be charged by input and output voltage difference V in − V out .Afterwards at instant t2, S2 is switched OFF, so its C oss is charged by one part of the output inductor current I L .Meanwhile, voltage across S1 drops when voltage across S2 increases, so S1 C oss is discharged by one part of I L , and its stored energy E oss is transferred to the device S2.After deadtime dt, which is supposed to be longer than S1 and S2 E oss transfer time, device S1 is switched ON.At this switching cycle, S2 switches OFF in hard switching mode at t2 and S1 switches ON in ZVS soft switching mode.Then, L is reversely charged by output voltage V out , so I L changes the direction.At instant t3, S1 is switched OFF in hard switching mode and like previous switching cycle, stored energy in C oss of S2 is transferred by one part of I L to C oss of S1.After deadtime, S2 is switched ON at ZVS soft switching mode and finally at instant t4, it is switched OFF at hard switching condition.
Based on the above control signals, I L current waveform is illustrated in Fig. 10b.For device S1, it only has turn-OFF switching loss at instant t 3 which is due to the overlapping of V DS and I D .It does not have any more turn-ON switching loss at instant   obtained waveform, when 1200 V SiC-MOSFET is switching at 300 V, its E oss is about 4.4 μJ.
It is compared in Fig. 12 measured E oss of all the aforementioned devices, which shows that E oss of all the devices is close to each other.The GaN-HEMT is evaluated at 300 V in soft switching mode, which shows more experimental results than original results presented by the authors in [8].This result confirms with device C oss values given in Fig. 4. By subtracting E oss from device turn-OFF energy given in Fig. 9, device switching losses in this switching mode are compared in Fig. 13, where it is shown that 1200 V SiC-MOSFET still has less switching losses than 650 V SiC-MOSFET and GaN device has less switching losses than SiC device in ZVS soft switching mode.
By combining all the above measurement results, it is shown that GaN device produces less switching loss in comparison with a 600 V or 1200 V SiC device with the same current rating and it is more suitable than SiC device to be applied in below 300 V energy conversion.Conduction loss and switching loss of SiC and GaN power semiconductor devices are compared in the paper.In order to compare losses of devices with the same power rating, a theoretical analysis is given, where it is shown that SiC power transistors specific ON-state resistance (m Ω × mm 2 ) will reduce half if device maximal blocking voltage decreases from 1200 to 600 V. Unlike device specific ON-state resistance, device specific capacitance value F mm 2 of 600 V device is four times bigger than 1200 V device.Due to SiC material low mobility, it is thus found that ON-state resistance R ON (mΩ) and inter-electrode capacitance (F) of 1200 V SiC device is smaller than 600 V SiC device if two devices have the same current rating, suggesting a lower conduction loss and switching loss of 1200 V SiC device.
In order to validate the theoretical analysis, static and dynamic characteristics of 1200 and 600 V SiC power transistors are at first compared with 600 V GaN-HEMT, in which it is shown that GaN-HEMT has smaller R ON and inter-electrode capacitances than SiC devices.Meanwhile, R ON and inter-electrode capacitances values of a 600 V SiC-MOSFET is even bigger than a 1200 V device in some voltage range.
Afterwards, switching energy E sw of those devices is compared in hard switching mode, where it is shown that GaN-HEMT has less E sw than all the SiC transistors (JFET, MOSFET) and 1200 V SiC-MOSFET has less E sw than 650 V SiC-MOSFET, which confirms the theoretical analysis.It is also shown in the results that 1200 V SiC-MOSFET and 650 V GaN-HEMT have bigger turn-ON E sw than turn-OFF E sw .
In order to reduce device turn-ON E sw , a ZVS circuit is used to evaluate devices in soft switching mode.Device output capacitance stored energy E oss can thus be measured, so device turn-OFF losses due to switching current and voltage overlapping can be obtained.By subtracting E oss , it is shown that GaN-HEMT has less switching losses than SiC devices in ZVS soft switching mode and 1200 V SiC-MOSFET still has less switching losses than 600 V SiC-MOSFET.
Based on all the experimental results, it can be concluded that 1200 V SiC-MOSFET has less switching losses than 600 V SiC-MOSFET in both hard and soft switching modes and GaN-HEMT is more suitable than SiC devices for vehicle-based power electronics systems in below 300 V electrical energy conversion.

Fig. 1
Fig. 1 Structure of different power transistors (a) MOSFET structure (b) JFET structure

Fig. 2
Fig. 2 Ideal switching waveforms and switching losses calculation

Fig. 3
Fig. 3 Device R ON comparison and measurement (a) Device RON comparison at 25 ∘ C (from datasheet), (b) Electrical circuit to measure R ON

Fig. 4 Fig. 5
Fig. 4 Inter-electrode capacitance comparison of different SiC and GaN devices

Fig. 6 Fig. 7
Fig. 6 Electrical circuit of the switching circuit and their realisation (a) Electrical circuit of the switching circuit, (b) Switching circuit of SiC-JFET, (c) Switching circuit of SiC-MOSFET, (d) Switching circuit of GaN-HEMT

Fig. 8
Fig. 8 Switching waveforms comparison when device switches at V DS = 300 V and I D = 10 A (a) Device ON, (b) Device OFF

( a )
Electrical circuit, (b) Control signals with load current waveform IET Electr.Syst.Transp.This is an open access article published by the IET under the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0/)

Fig. 11
Fig. 11 Measured 1200 V SiC-MOSFET switching waveforms and E oss when it switches at 300 V/15 A in soft switching

( a )Fig. 12
Fig. 12 Device stored energy E oss comparison of different SiC and GaN devices