Posted on 29 June 2019

Applying High Frequency GaN Transistors to Motor Drives

Free Bodo's Power Magazines!




No Diodes? Really?

The low charge associated with gallium nitride high-electron-mobility transistors (GaN HEMTs), and the consequent low switching loss and fast switching speed, make these devices attractive for any switch-mode power application. A lesser known characteristic, the lack of a parasitic junction, is also particularly advantageous for motor drives, in that it relieves the need for freewheeling diodes.

By Jim Honea, Transphorm Inc., Goleta, CA


This article will explain how the diode-free operation is possible, and then present a methodology for using GaN to advantage in motor drives, including a presentation of efficiency data achieved in motor testing with a GaN three-phase bridge.

Diode-Free Bridges

The typical three-phase bridge used in AC motor drives is shown in Figure 1 (a). Each switch (shown here as an IGBT) is clearly paired with a freewheeling diode, which, for voltages beyond about 200V, will be either a silicon junction diode, or a GaN or SiC Schottky diode. Some transistor technologies include junctions which could, if permitted, serve the function of the freewheeling diode, the body diode of a MOSFET being one example. Figure 1 (b) shows a threephase bridge constructed with six GaN HEMTs, and no diodes [1].

Three phase bridges made with (a) IGBTs and discrete diodes, and (b) GaN HEMTs.

When presented with a schematic of this sort, circuit designers will typically ask if this doesn’t merely indicate that the body diode of the GaN transistor is acting as the freewheeling diode. The answer is no, because there is no body diode. This is true, almost by definition. High electron mobility is achieved in a HEMT precisely because the sheet of mobile electrons forms in a region of pure, undoped crystal. If there is no doping, there can be no junction between p and n doped regions. The GaN material is truly insulating everywhere except in the channel, but the electrons in the channel are free to flow in either direction. Therefore, while the device will block appreciable voltage in only one direction (Vds>0), it can conduct current in either direction through the channel. The difference between this and conduction through the body diode of a MOSFET is profound. Even though the freewheeling current will flow through the same source and drain pins in a MOSFET, the path it takes in the die is completely different for the two directions: the forward current is a flow of majority carriers only, down the channel and across the drift region, while the reverse current is a flow of minority carriers, injected across a barrier. This injected charge has to be recovered when the voltage reverses, constituting the troublesome reverse recovery charge (Qrr). To emphasize the difference, Figure 2 contrasts Qrr for a 600 volt GaN HEMT and a Si MOSFET of similar ratings.

The device tested in Figure 2 was actually a hybrid device, in which a low-voltage, normally-off silicon MOSFET is coupled with a high-voltage, normally-on GaN HEMT. The configuration of the composite hybrid device, shown in Figure 3 (a), along with the symbol 3 (b), is such that the low-voltage silicon device provides the desired normally-off functionality, while the GaN device provides the advantages of low on-resistance, low charge, and high voltage blocking. This raises the question concerning whether this device can be used in a bridge application and still have the diode-free characteristic, since, evidently, the free-wheeling current will flow in the body diode of the silicon device. In fact, the benefit is nearly the same. This is because the silicon device is a low-voltage device (20-30V). The low blocking voltage permits a short drift region, and therefore a small volume in which injected charge will be stored.

Reverse-recovery charge test result  (a) a fast-recovery CoolMOS and (b) a Transphorm GaN HEMT

Schematic (a) and symbol (b) for a hybrid transistor comprising a low-voltage normally-on silicon MOSFET

The voltage drop across the GaN switch, whether single die or hybrid, can be reduced during the reverse conduction period by driving the gate to enhance the channel. Figure 4 shows gate-drive waveforms to accomplish this, assuming Vgs1 is the gate-source voltage of the actively switched transistor, and Vgs2 is the gate-source voltage of the transistor carrying the freewheel current. The freewheel current flows in times A and B, but the freewheel device is only enhanced during time B. During time B, the voltage drop is just Id *rds-on. A significant point is that, because Qrr is low, there is no great penalty in providing plenty of safety margin in the the dead times (designated as time A in the figure). During these dead times, the voltage drop is increased by the forward voltage of the silicon MOSFET’s body diode, in the case of the hybrid, or the threshhold voltage (plus overvoltage, depending on the current) in the case of the enhancement-mode HEMT.

Gate-drive waveforms to lower conduction loss during the reverse conduction period.

High Frequency Motor Drive

Having explained the operation of a diodefree, three-phase GaN bridge, we will consider how to take advantage of the simple design and potential for low loss in a motordrive application, beginning with some background on the problem.

A logical first step towards improving overall system efficiency in any motor system is to use an inverter to power the motor, with the subsequent benefit of variable speed operation. However, application of inverter drives to motor systems introduces additional stresses which has necessitated improvements in motor design [2,3]. A key point is that these improvements – measures such as improving the magnet wire insulation and providing static discharge paths – enable realization of the benefits of variable speed operation, but do not actually improve the intrinsic power efficiency of the motor itself. On the contrary, some loss of motor power efficiency is the unavoidable consequence of PWM operation associated with using an inverter. This loss of efficiency may be attributed to two basic mechanisms: additional real power converted to heat in the motor itself due to nonproductive harmonics in the current waveforms, and loss which occurs in the inverter due to reactive elements in the motor and wiring.

Any real power delivered to a motor at a frequency other than the fundamental (that of the back EMF) only serves to heat the motor. The higher the switching frequency, the lower the distortion which may be achieved, including ripple current at the switching frequency. This principle is applied in other applications, such as powerfactor correction circuits, where 100kHz (and higher) switching is common in order to reduce harmonics on the 50/60Hz power lines. The obvious conclusion, therefore, seems to be that motors should also be driven at switching frequencies above 100 kHz; but this provides additional complications. Regarding efficiency, the most immediate problem with increasing switching frequency is increased switching loss.

For a motor drive built with high-speed switching devices, such as the GaN HEMT’s discussed here, the dominant part of the switching loss is most likely contributed by external capacitance. The energy of a linear capacitor, charged to voltage V, is ½CV2. This energy is lost when the capacitance discharges in a hard-switching transient. In fact, if the node is truly hard switched in both directions, the same energy is lost in the charging process, and the energy loss per cycle is now CV2. For now, we will assume that either the charge or discharge transient is soft, meaning the energy is transferred from the total inductance where it has been temporarily stored during the previous cycle. In this case, the loss per switching cycle, per node, is ½ CV2, and the total switching loss due to external capacitance in a three-phase drive is 3(½CV2)fs. Note that this is independent of the rise and fall times: the energy is placed on the capacitors, and then dumped to the heat sink. The presence of an EMI output filter, which simply slows the rise and fall, does not lower the loss.

Conceptual schematic showing external capacitance connected to a switching node

The motor also has a number of capacitances between its windings and the motor frame, and between the individual phase windings themselves. For the 1Hp AC induction motor used in the following tests, capacitance values on the order of nanofarads per phase are probably correct [4].

A solution to the problem of capacitive switching loss is to isolate the external capacitances from the switching signals. An output filter designed to eliminate the switching frequency will accomplish this, as illustrated in Figure 6.

Using a filter to isolate the inverter from external capacitances

Of course, the inductors used in a filter also have winding capacitance, and will also contribute core loss. But these losses can be minimized through design trade offs which are not necessarily available in the motor and cabling. The motor’s winding has a primary purpose of transferring energy to the rotating load, and minimizing the associated capacitance and core loss can only be done within constraints imposed by that primary purpose. Working against this factor is the low switching frequency of existing inverters. A typical filter that is effective at 8kHz must use large, relatively lossy inductors. With a higher PWM frequency, the filter can be better optimized for both efficiency and cost or size. This consideration leads to what may be the ideal solution – a high frequency inverter, with low inherent switching loss, that is isolated from the motor and cable by an output filter.

A concern that arises when fast rise and fall times are discussed is electromagnetic interference (EMI). As pointed out by Carsten and Mammano [5], three elements must be present for EMI to take place - a source, a coupling means, and a receptor (or victim, as they put it). To the extent the coupling means can be minimized, the severity of the source can increase. The charge on the switching node produces an electric field that terminates on some other conductor in the system, which is to say, a parasitic capacitance exists between the nodes. If the switching signal is considered the EMI source, then this capacitance is the coupling means. The presence of the electric field itself is not typically a problem. What is a problem is the current produced when the switching node slews, I = CdV/dt. This current often finds its way to the chassis, or ground return, constituting a common-mode current. However, if the capacitance between the switching node – the source, where the high dV/dt exists – and any other nodes in the circuit approaches zero, then the slew rate may approach infinity. This ideal cannot be appreciably realized if the motor is driven directly with the switch-mode signal, but may be if the only connections are two transistors and one lead of a small inductor, internal to the inverter, as indicated in Figure 7. One final observation is that the use of a properly shielded cable can limit emissions, but most likely increases total capacitance.

Minimizing the area indicated minimizes the parasitic capacitance, which is a coupling means for EMI signals

GaN Motor Drive Test Results

Figure 8 shows the basic design of a simple inverter built with a 6-transistor module, and associated test circuit. The filters used are simple LC filters, with 200µH inductors and 0.3µF capacitors. A simple V/F, open-loop algorithm was used in all cases.

Block diagram of test circuit for testing inverter-drive efficiency

The efficiency of this system driving a variable resistive load up to 2Hp is shown in Figure 9. Figure 10 shows the result when driving a 1Hp AC induction motor (Marathon Electric, Micro Max series) with variable torque load.

Electrical efficiency - Pac(out)/Pdc(in) -of the GaN inverter driving a 3-phase resistive load

Electrical efficiency of the GaN inverter - Pac(out)/Pdc(in) - driving a 1Hp AC induction motor with variable torque load

A comparison was made between the GaN inverter and a 230VAC/2Hp commercial inverter drive to demonstrate the impact on the efficiency of the motor itself. Figure 11 shows the result, confirming that the motor itself operates less efficiently when driven directly with PWM signals. Here, the ratio of mechanical power to three-phase electrical power is the plotted quantity. The commercial inverter was operated at 16 kHz, directly connected to the motor (without filter). Note that the additional switching loss due to the cable and motor capacitance is not included here, since that energy is dissipated in the inverter, and only appears in the motor as reactive power.

Electromechanical efficiency of the 1Hp ACIM motor when driven by a GaN inverter, and by a commercial, silicon-based drive

A point to be acknowledged is that the openloop drive algorithm used here lends itself quite easily to a drive with filtered outputs. More sophisticated algorithms which require sensing of phase currents and voltages could need some adaptation for use with filtered outputs.


GaN power transistors can simplify bridge circuits for motor drives through elimination of freewheeling diodes. They can also enable increased efficiency through high-frequency switching, although the method for realizing that benefit is complicated by the nature of the load. We have presented a methodology for achieving both benefits through addition of small output filters, keeping the high switching frequency isolated from the motor and wiring. Test data confirms that high efficiency can be achieved in the inverter itself, even with the filter, while the system-level efficiency is improved by reduction of core loss, copper loss, and capacitive switching loss.



1) J. Honea, Y Wu, 2011, Bridge Circuits and Their Components, United States Patent 7,965,126 B2.
2) S Bell, J Sung, “Will your motor insulation survive a new adjustable frequency drive?”, IEEE Transactions on Industry Applications, Volume 33 Issue 5 (Sep-Oct 1997) Pages 1307-1311.
3) Bearing currents in modern AC drive systems, ABB Technical Guide No. 5
4) A. Muetze, A. Binder, “Calculation of motor capacitances for prediction of the voltage across the bearings in machines of inverter-based drive systems”, IEEE Transactions on Industry Applications ,Volume: 43 Issue: 3 (May June 2007) Pages: 665-72.
5) B Carsten, B Mammano, “Understanding and Optimizing Electromagnetic Compatibility in Switchmode Power Supplies”,



VN:F [1.9.17_1161]
Rating: 0.0/6 (0 votes cast)

This post was written by:

- who has written 791 posts on PowerGuru - Power Electronics Information Portal.

Contact the author

Leave a Response

You must be logged in to post a comment.