The availability of Gallium Nitride (GaN) devices for power conversion is improving, with many manufacturers coming out with catalog parts. The application of these devices has gained the keen attention of power-system designers from all over the world. Although a lot of study has gone into understanding their properties, the selection of GaN devices for various power-conversion applications based on their properties remains not well understood.
After much study dedicated to GaN as an alternative to silicon in power switching, multiple manufacturers now offer GaN switching devices for power-conversion applications. However, you must look at the properties of GaN devices in detail before assessing their suitability and advantages.
Device Architecture
GaN switching devices come in two different types based on their internal architecture: enhancement mode (e-GaN) and cascoded depletion mode (d-GaN). An e-GaN switch operates like a normal silicon metal-oxide semiconductor field-effect transistor (MOSFET), although it has reduced gate-to-source voltage levels. An e-GaN device also has a simpler architecture and packaging, low on-resistance, and zero body-diode reverse recovery (there is no body diode, but the channel itself is bidirectional in nature and behaves like a body diode).
The first (and main) concern with this type of device is the critical nature of its gate-drive design. The problem is that the device’s fully enhanced gate-drive voltage is very close to its breakdown voltage—the safety margin is typically only about 1 V. This might cause a device failure in the event of a voltage spike or parasitic ringing. Second, the comparatively lower gate threshold voltage could reduce noise margins. A third concern for these devices—although not very serious—is the higher gate-leakage current, which could increase gate-driver dissipation.
The depletion mode GaN device offers both performance as well as manufacturing advantages. Its normally “on” nature may be a problem during power-up and other abnormal operating conditions, however. It also requires the use of a negative supply. You can overcome this problem by connecting the depletion mode GaN high-electron mobility transistor (HEMT) in series with a low-voltage silicon MOSFET in the cascoded d-GaN structure. The gate of the HEMT is shorted to the source of the MOSFET, while the HEMT source connects to the drain of the MOSFET. As Figure 1 shows, the gate-to-source voltage of the HEMT is the source-to-drain voltage of the MOSFET. So the silicon MOSFET can control the turning on and off of the GaN HEMT.
The main advantage of this structure is that the complete cascoded d-GaN switch has the gate characteristics of a low-voltage silicon MOSFET. Therefore, existing commercial MOSFET gate drivers can easily drive the cascoded d-GaN switch. There are no unknowns to deal with, since the gate characteristics of silicon MOSFETs are well-known.
You will have to make some compromises in overall performance because of the additional series silicon switch. The most significant impact on performance is arguably due to the reverse recovery associated with the body diode of the silicon MOSFET. Because the cascoded d-GaN switch is a series combination, it will have reverse recovery while conducting in a reverse direction, unlike an e-GaN switch. The next significant effect is the possibility of the silicon MOSFET avalanching during turn-off due to the charge imbalance between the drain-to-source capacitances of the two series devices. This can potentially increase switching losses and decrease reliability.
A cascoded d-GaN has increased packaging complexity and cost due to the additional series silicon switch. The higher number of devices and interconnections increase issues related to reliability. Parasitic inductance and capacitance between the silicon switch and GaN HEMT may cause delay and oscillation during switching transients and impact electrical performance.
Another unfavorable effect with d-GaN devices is the increased on-resistance because of the addition of the on-resistance of the silicon MOSFET. The increase can be significant for lower-voltage (<200 V) cascoded d-GaN devices. So for low voltages, e-GaN switches are a better choice. For a high-voltage (600 V) cascoded d-GaN device, the additional resistance may be only about 5 percent of the overall on-resistance; at this voltage level, cascoded d-GaN is still a viable option.
TI Direct-Drive Gan
After comparing e-GaN and d-GaN structures, it is obvious that most of the issues associated with the cascoded d-GaN are from the simultaneous switching of the silicon MOSFET and GaN HEMT. TI’s direct-drive technology overcomes this problem by using only the silicon MOSFET to enable the device. Figure 2 compares the configuration.
As mentioned earlier, in cascode GaN, both the silicon and GaN devices turn on and off together. However, in the TI direct-drive GaN, the silicon MOSFET is only used to overcome the problem of power-up shoot-through. The silicon MOSFET is not switching; it turns on during power-up and stays on until power-down with an enable signal that’s applied only when the negative bias voltage to turn off the GaN HEMT is available. Integrating the negative voltage supply and gate drives along with the power switches ensures reliable and precise control over switching of both the silicon MOSFET and GaN HEMT. Driving the GaN gate directly reduces the gate charge significantly. Completely eliminating the body diode’s reverse recovery and silicon-switch avalanching results in a considerable reduction in switching losses.
Other advantages resulting from the integration of the drive circuit into the power devices include the ability to control the switching slew rate, cycle-by-cycle overcurrent protection by sensing the voltage drop across the silicon MOSFET and over-temperature protection.
To Soft Switch Or Not
There is a general perception among many power system designers that GaN power switches can make soft or resonant switching topologies irrelevant due to their ability to switch very fast. The assumption is that fast switching capability can achieve comparable or even better efficiencies with hard-switching topologies; thus, soft switching may become irrelevant. While this is true at the switching frequencies currently achievable with silicon MOSFETs, there is significant EOSS loss associated with GaN devices. As the switching frequency increases, the EOSS loss becomes the most significant loss component. Since the stored energy in the output capacitance of GaN devices is much more easily recoverable compared to silicon super-junction MOSFETs, it makes a lot of sense to go for soft or resonant switching, especially at multi-megahertz switching frequencies.
Conclusion
The device architecture and performance parameters of d-GaN and e-GaN devices continue to evolve as efficient, robust, and cost-effective alternatives to silicon MOSFETs in power conversion. No doubt, many of the performance parameters discussed in this article will be improved significantly in the future. Higher levels of integration of GaN devices like half bridge devices and integrated gate driver with protection circuits will make their use in power conversion far easier. The adoption of GaN devices in soft switching topologies can push the power density of power converters to levels unprecedented with silicon switching technology.
