GaN Modules Boost Switching Frequencies While Maintaining Efficiency

In 2012, we finally got real insight into gallium-nitride (GaN) power devices. Efficient Power Conversion (ePC) published a 200-page textbook about the devices.¹ If you want to understand how to compare figures of merit for GaN and silicon devices, for example, it’s there on page 24. International Rectifier (IR), meanwhile, disclosed the architecture of its first GaN products, the IP2010 and 2011 Integrated Power Stage modules.

In terms of product count, ePC’s seven enhancement-mode high-electron mobility transistor (HEMT) devices outnumber IR’s two modules, which are based on the company’s depletion-mode devices. On the other hand, the IR modules have more pizzazz. They’re drop-in solutions for switching power supplies. Just add a pulse-width modulation (PWM) stage to drive them.

IR’s “Cascode” Topology

The IR modules are configured with the GaN device as the upper switch and a conventional silicon MOSFET as the lower switch (Fig. 1). This arrangement is identical to the cascode amplifier topology (in bipolar junction transistor terms) in which a high input-impedance common-base stage drives a low output-impedance common-emitter stage. It provides all the performance of a common-emitter stage with a much smaller Miller effect and much higher output resistance.

1. International Rectifier’s IP2010 and 2011 modules provide the upper and lower switches of a switched-mode power supply, with a GaN device as the upper switch and a conventional silicon MOSFET as the lower. Viewing this configuration as an adaptation of the classic cascode amplifier structure provides insights into the design’s cleverness.

The cascode was first developed to get better high-frequency performance in audio applications, and the higher output resistance was viewed as a bonus. Now designers take advantage of both features in a variety of situations.

Comparing Turn-On Behavior

The turn-on event of a standalone standard MOSFET can be compared to the same turn-on in IR’s GaN cascoded device.² The stages that a conventional MOSFET goes through as it turns on will be familiar to anybody who has worked with power devices. As the gate-source voltage V_gs initially ramps up from zero, nothing happens until it reaches the device’s threshold voltage at time t₁. At that point, drain current I_d begins to flow. Up until this point, the voltage across the MOSFET has not changed.

From the threshold-crossing event to the onset of the Miller plateau (time t₁ to t₂), charge builds up across the gate-source capacitance and drain current I_d starts ramping up. When I_d hits the Miller plateau at time t₂, the drain current goes flat. Until this time, drain-source voltage V_ds was constant, but now it begins to fall rapidly. By time t₃, at the end of the Miller plateau, with a full charge on C_gs, V_ds is equal to I_d x R_DS(on), and the MOSFET is saturated. Most of the MOSFET’s switching loss arises from the time spent traversing the Miller plateau.

IR’s GaN-FET cascode module is different (Fig. 2). The GaN device is the upper switch in a half-bridge. The lower device is a conventional silicon MOSFET. From t₀ to t₁, V_gs again ramps up to the threshold and then flattens out across the Miller plateau of the lower MOSFET switch until time t₂. At this point, the V_ds of that lower MOSFET is still high, so the GaN device remains in the OFF condition.

2. Looking at the switching waveforms for the cascoded device, one can see a very short Miller plateau and the level-shifting effect of the upper GaN device. Because the low-side element is a low-voltage MOSFET, the ON-state performance of the cascode sees only a small penalty added for the added element. The contribution of the low-side device is kept to less than 10% of R_DS(on) in most cases. Similar advantages apply during turn-off.

During the time it takes to traverse the Miller plateau, the drain-source voltage of the MOSFET starts to fall. (The V_ds of the lower MOSFET is the negative of the fate-source voltage of the upper GaN device!) At t₃, the V_ds of the MOSFET reaches the threshold of the GaN device, which, being a high-transconductance device, turns ON very quickly (between times t₃ and t₄ in Figure 2).

If that explanation was a little difficult to follow, consider the upper GaN device as a common-gate amplifier, essentially a level shifter. As the lower MOSFET turns ON, the upper device actively collapses its own drain-source voltage. Because it is a GaN device, this happens more rapidly than it would in a conventional MOSFET, because its capacitances and total gate charges are a lot smaller than in an equivalent superjunction MOSFET.

Getting back to the cascode analogy, the classic cascode topology was developed to meet a need for greater speed from a vacuum tube, and later, a bipolar junction transistor (BJT) or FET. With a cascode arrangement, the upper active element serves to reduce the Miller capacitance of the lower element by actively swinging in the desired direction. In a BJT application, one could look at the upper element as a common-base level-shifter that added speed. Switching circuits are similar in this regard, only we seek to spend minimal time in the active region.

When the intrinsic body diode in the IR module’s low-side conventional MOSFET conducts, the gate and source terminals of the upper GaN device rise above the (negative) threshold voltage and into the saturated ON state. The GaN device is saturated. Thus, the overall freewheel current path is through the lower MSOFET’s intrinsic body diode and the channel of the upper GaN device.

The intrinsic diode in the low-side, low-voltage MOSFET is much faster than a high-voltage MOSFET’s body diode would be. It also has a much lower V_f and reverse-recovery characteristic. The added I_d x R_DS(on) voltage of the GaN device raises this “total V_f,” but not as much as a silicon high-voltage PN-junction device would. The result is a high-voltage switch with an antiparallel rectifier that rivals silicon-carbide (SiC) diodes in performance.

According to International Rectifier, the IP2010 and IP2011 modules allow switched-mode power-supply designers to build switching converters at five to 10 times the switching frequencies that are presently used while maintaining the same efficiency.