Combining GaN and SiC for cost-effective power conversion
There are practical difficulties with using wide bandgap devices to make cost-effective and reliable products. For the best overall result, OEMs should consider combinations of the technologies, even including silicon in the mix. This article evaluates some of the issues and solutions.
Wide bandgap devices have some limitations
Make no mistake, SiC and GaN can outperform silicon, but they do have limitations and discrete devices are currently more expensive than their silicon counterparts. There are operating conditions when they do outperform silicon, and where SiC is better than GaN (or vice versa). Knowing why become important considerations.
Some headline differences between Si, SiC and GaN are shown.
Voltage rating is a major consideration
Voltage rating is a significant difference between wide bandgap (WBG) devices currently available. SiC can be rated to at least 1200V whereas GaN currently tops out at 650V. Worse, GaN breaks down catastrophically with over-voltage whereas a SiC-MOSFET has an over-voltage avalanche rating and will survive up to a specified energy level.
It’s not quite as bad as it sounds, as GaN manufacturers allow a significant margin above their rated value. They won’t guarantee the figure though, so a prudent engineer will not operate them anywhere near 650V. This limits the applications in which GaN can be safely used. For example, the common situation of generating a DC link voltage from rectified three-phase 400VAC produces over 560VDC.
With tolerances on top, this is too close to a GaN device rating used in a downstream power conversion stage. This assumes a converter topology that limits switch voltages to the DC link value. Other topologies operating even from just single-phase mains can also produce high peak voltages. For example, the humble flyback used in a cell phone charger can generate more than 500V stress on the switch at high mains, with leakage inductance spikes that go higher still. GaN can be used in this application to achieve high efficiency and small size, but only typically with the switch co-packaged in a power and control IC to achieve precise gate drive and voltage overshoot control.
The voltage limit of GaN stems from its construction. Current devices are lateral cells that break down across their surface, so a higher voltage rating means a larger die area per cell. GaN wafers are already expensive and smaller than Si and SiC, so decreasing yield with a larger die is not attractive commercially. Work is underway to fabricate vertical, trench-style GaN MOSFETs to overcome this limitation, but they are not yet in economic production.
At power levels higher than seen in cell phone chargers, the switch is usually a separate discrete device for dissipation reasons and then obtaining reliable operation with GaN becomes more difficult, with interconnections requiring RF-style considerations. In most cases, edge rates of both GaN and SiC switches have to be deliberately slowed for stable operation and to minimize electromagnetic interference (EMI). This reduces efficiency and offsets the advantage of using WBG devices.
The body diode effect can be critical to the application
Another characteristic that affects practical performance is the way that switches conduct in reverse, the so-called third-quadrant operation. This naturally happens in hard-switched converter topologies with inductive loads and in motor drives, for example.
Si- and SiC-MOSFETs conduct in reverse through a parasitic body diode and commutation before the channel is actively turned on. This diode has a high forward voltage drop, along with stored charge which causes a spike in dissipation when released. SiC has a much higher forward drop than silicon, but less stored charge. GaN has no parasitic diode and stored charge effect but does commutate through voltage gradients in the device turning the channel on for reverse conduction. The drop can be several volts, however – the sum of the turn-on threshold voltage and any negative turn-off voltage applied.
Whether the reverse recovery current losses and forward voltage drop of the body diode effect are significant depends on the circuit topology. For example, in soft-switched, resonant converters such as the LLC, the absence of recovery current is less important and GaN only wins because of its fast edge rates and low device capacitances, giving low dynamic losses at high frequency. In a hard-switched converter, GaN’s lack of body diode recovery is a major plus, but if the frequency is increased to take advantage of this, the proportion of any dead time to cycle time increases and dissipation in the high voltage drop of the body diode effect starts to be problematic.
There are clearly basic trade-offs to consider when choosing SiC or GaN, and many other factors come into play as well, including the variation of parameters with temperature, complexity, criticality of gate drive, available packaging, and not least, device cost. Even though this is not straightforward, GaN is currently more expensive but if it needs less or no heatsinking and other related components such as magnetics, can be smaller and cheaper due to its high switching frequency, then perhaps the overall system cost is lower.
An example hybrid application
To illustrate the choices possible, we could consider a common three-phase motor drive circuit, powered from 230VAC nominal with a requirement to have very high efficiency at around 1kW. Power factor correction is necessary and a high-efficiency solution is the totem-pole PFC stage. However, above a few hundred watts, this topology needs to be operated in hard-switched, continuous-conduction mode to minimize stress levels. The body diode recovery currents in Si- and SiC-MOSFETs limit efficiency and therefore GaN is the choice for the lowest losses, but with a device cost penalty. However, the switching frequency can be several hundred kHz while maintaining high efficiency, allowing a smaller and lower-cost inductor in compensation. The highest voltages seen are normally around 400V, so there is sufficient margin, even with 600V-rated GaN devices. Si-MOSFETs are shown as the slow rectifiers in the PFC stage as they switch at the mains frequency and have minimal dynamic losses. These could be simple diodes at a lower cost for a small efficiency penalty.
In the inverter section driving the motor windings, commutation occurs due to the inductive loads, with potential third-quadrant reverse recovery losses. However, there is little point in operating at a high switching frequency as the magnetic element is the motor winding which scales in size with the application, not pulse-width-modulation (PWM) frequency. Additionally, high frequency exacerbates EMI problems with the typical long wire connections to the motor. This means that the motor drive operates usually well below 20kHz with correspondingly fewer reverse recovery losses per second and low resultant dissipation. In this case, a SiC-MOSFET can be almost as efficient as and lower cost than GaN. SiC is also more robust against the fault currents from inevitable motor overloads and locked rotors that occur in real life.
Conclusion
While there is a continued price disparity between SiC and GaN switches, the case for one or the other in power conversion will depend on the application and wider system considerations, including energy savings over time from even decimal point savings in efficiency. Taking a holistic view, all technology options should therefore be explored to achieve an optimum solution.
