Making GaN Power Electronics Universal

As GaN power electronics is gaining momentum, an interesting question to ask is whether GaN will remain a niche player like GaAs or whether it will truly displace silicon power MOSFETs. Based on the experience of how bipolar power transistors were displaced by power MOSFET in the 1980s, several aspects are important.

The first is that using GaN must be much more beneficial and it must enable many new applications to make it worthwhile. The second is that it must be easy to use and universally available in a wide range of current and voltage ratings. Finally, it must be very cost-effective and highly reliable.

This article will focus on how GaNPower International Inc., a Canada based company offering advanced GaN power devices and system solutions, is able to make GaN easier to use and more universal.

 

Pushing the Breakdown Voltage Higher

It was commonly believed that high voltage applications (> 1000V) were exclusive domains of silicon IGBT and SiC (both vertical devices) until recently GaNPower proved otherwise. In August 2018, GaNPower International Inc. announced [1] a major breakthrough. By innovative design, GaNPower had succeeded in the tapeout of its first commercial lateral single-die E-mode GaN transistor suitable for rating at 1200V breakdown voltage. The breakthrough was supported by on-wafer testing as indicated in Figure 1 and the typical threshold voltage was found to be around 1.4 volt (E-mode)

Understandably, device packaging for high voltage lateral devices was challenging. So far GPI has just released GaN E-mode devices rated at 1000V and 1100V on TO252 (15A/95mOhm) [1].

 

Figure 1: On-wafer and after packaging results for high voltage Emode GaN.

 

With continued improvement in packaging, it is certain that single-die E-mode devices will soon be commercially available at ratings of 1200V and higher. It is worth noting that increasing the breakdown voltage of GaN does not mean the significant sacrifice of switching performance. The Qg*Rdson product of 650V/15A GaN from GaN-Power is around 300 (mOhm*nC) which is to be compared with that of 1100V/15A GaN at 320 (mOhm*nC). It is a pleasant surprise that high voltage lateral GaN works so well.

 

Making the Gate Easier to Drive

Behaviors of E-mode GaN transistors are very similar to those of conventional power MOSFETs (except they are much faster). Therefore power engineers can use their system experience with minimal additional training. However, special attention should be paid to how the gate is driven. Taking the popular p-GaN GaN/silicon as an example. The Schottky barrier associated with the gate would be driven into exponential current increase when gate voltage reaches 13-14 volts (see simulation and experimental data in Figure 2).

 

Figure 2: On-wafer results of GaN integrated with over-voltage-protection IC. Simulated gate current for a reference device is also down.

 

Due to the above possible catastrophic gate damage, extreme care must be exercised in designing and implementing the circuits such that no instantaneous over-voltage would happen to the gate.

Even picohenries (pH) of parasitic inductance can cause several volts of overshoot in gate drive and power loops and this can cause irreversible damages to the device. The above requirement in design and implementation is easier said than done and this continues to be a major hold back in the adoption of GaN.

To make the GaN device easier to drive, GaNPower invented a GaN specific over-voltage protection (OVP) circuitry monolithically integrated with GaN power device. The OVP circuit uses the gate voltage as its auxiliary power supply and the whole new IC appears to be just another GaN discrete power device except the gate would not easily be damaged. Preliminary on-wafer testing results (Figure 2) are very promising while the packaging effort is on-going. The gate current is clamped at a controlled value when the input voltage is above the desired driving voltage (6V).

The over-voltage protection can, in theory, go up to 100V without damaging the device thus providing effective protection. The benefit is that with little or no degradation in switching performance and a slight increase in semiconductor wafer area, the GaN works just like a silicon MOSFET without running the risk of the catastrophic gate damage. System engineers would be more willing to use such power device/IC since they behave like MOSFET except it has much better performance.

 

Next Generation GaN-Specific System Solutions

The rationale for using GaN has been that since it is still a relatively expensive device, it should be used where high switching frequency and high power density are required. With patented circuit topology and control method [2], GaNPower is creating a series of next-generation system solutions specific to GaN technology.

It is generally agreed in the power industry that GaN is especially suitable for resonant topologies at the high switching frequency. The reason is a smaller Qg*Rdson enabling faster charging/discharging of the internal capacitors of GaN transistor and therefore a much smaller dead time is required in switching control.

Conventional LLC resonant circuits have a major shortcoming: they are difficult to parallelize (multi-phasing or multi-staging). The reason is that the voltage gain of LLC converter is sensitive to the resonant frequency, or resonant inductor and resonant capacitor values. When connected in parallel with same switching frequency operation, any slight difference in L and C value due to tolerance will cause current imbalance and therefore, current sharing cannot be achieved. In other words, one phase will deliver a majority of the current and the other phase deliver a small current. So all the current application of LLC and other resonant converters uses a single converter to deliver all the power. This limits the output power carrying capability of the resonant converters. This problem is especially true for low voltage (such as 12V, 14V), high output current (such as 100A, 200A) application where the conduction loss will be very high and therefore, reduces the efficiency and increases the cooling requirement. The transformer size is also increased, which defeats the purpose of using GaN for high density. This problem is more severe for the EV onboard DC-DC converter with 14V output voltage and 140 – 280A load current.

Combined with the fact that the GaN device is a lateral power device which is very difficult to be designed to carry high current, due to packaging limitation. With larger area lateral device, it would be difficult to wire bond (or other connection means) the center area of a large device without causing the significant increase in on-resistance and issues in wire crossing.

GaNPower proposed a Switch-Controlled-Capacitor (SCC) technology to solve the current sharing problem of the LLC resonant converters with two or more LLC converters connected in parallel [2]. By controlling the equivalent resonant capacitor value, the resonant frequency of each LLC phase can be made equal even with L and C tolerance. Therefore, perfect current sharing can be achieved among all the phases. The extra cost for SCC could be as low as one MOSFET with the source connected to primary ground, which significantly simplifies the gate drive. The MOSFET operates at zero-voltage-switching (ZVS) condition.

When phase-shedding (with some phases shutting down) operation is used, peak efficiency operation can be maintained over a very wide output current range. As a proof of concept, a 600W (2x300W) two-phase LLC DC-DC converter was constructed (Figure 3).

Good efficiency and phase shedding were achieved. With the SCC technology for multi-phasing, high-frequency switching, high power density, and high efficiency would be achieved to maximize the benefit of using GaN for any high power conversion systems.

 

Figure 3: Proof of concept demo of a 600W(2x300W) switch-controlled capacitor LLC DC to DC converter.

 

The team at GaNPower also demonstrated that LLC topology is not only suitable for GaN in medium to high power applications, it is also a good choice for GaN in lower power application such as 65W power adapter. Table I list a comparison between ACF and LLC and it is clear that LLC is fundamentally more suitable for commercialization of GaN in 65W power rating. A recent prototype of the GaN-LLC design of the 65W power adapter is shown in Figure 4 and GaNPower is expected to make it available commercially soon.

 

Figure 4: Demo systems of 65WPD using LLC topology and GaN at 1MHz switching.

Table 1: Comparison of ACF and LLC for 65WPD.

 

With innovative GaN device designs and next-generation LLC solutions, GaN technology appears to be increasingly likely to displace silicon power MOSFET and SiC in the near future. The author acknowledges helpful comments from Drs. Gary Dolny, Yanfei Liu and Yue Fu.

 

About the Author

Dr. Simon Li obtained his Ph.D. from the University of British Columbia through the prestigious CUSPEA (China-U.S. Physics Examination and Application) program in the early 1980th. He was previously with the National Research Council of Canada (NRCC), where he developed the world’s first commercialized laser diode simulator. He founded Crosslight Software, Inc. in 1995 as a spin-off from the NRCC which has become a leading provider of TCAD tools for the semiconductor industry. Dr. Li is a well-known expert in semiconductor device physics and a semiconductor industry veteran with a rich background in high-tech company management.

 

This article originally appeared in the Bodo’s Power Systems magazine.