Devices that control electrical energy as powerful microsecond pulses, such as power grid protection equipment, fusion experiments, and high-energy accelerators, need switching elements with high voltage tolerance and the ability to withstand large currents. Silicon (Si) has long been used in these fields, but because of performance limits rooted in its material properties, the industry is shifting toward wide-bandgap semiconductors such as silicon carbide (SiC) and gallium nitride (GaN).
Research is now moving on to the next generation of materials: ultra-wide-bandgap (UWBG) semiconductors, with bandgaps above 4.5 eV. One leading candidate is gallium oxide (). In general, the wider a semiconductor's bandgap, the higher its breakdown electric field tends to be, so a thinner semiconductor layer may be able to block a higher voltage. Gallium oxide, however, has a major weakness. Its thermal conductivity is low, at 11 to 27 W/(m·K), so it is hard to carry heat generated inside the device out to the surroundings. Even with high voltage tolerance, passing a large current makes the device temperature rise sharply, and as a result the switching power capacity demonstrated in actual circuits has remained relatively modest.
An international research team led by the University of Hong Kong, with participation from Virginia Tech, Nanjing University, and others, took on this problem with "co-design," which optimizes the semiconductor device and its package as a single system. The result was a module of six gallium oxide diodes connected in parallel, which demonstrated short-pulse switching at 1,000 V and 1,000 A. The findings were published in the journal Nature Communications on March 30, 2026. According to the team, the switching power capacity achieved exceeds previously reported circuit operation with ultra-wide-bandgap semiconductors by more than two orders of magnitude.
Note, however, that the "1 MW" here refers to a "switching power capacity," defined as the product of the voltage that can be blocked in the off state (1,000 V) and the current that can be carried in the on state (1,000 A). The team did not build a power converter that can supply 1 MW continuously, nor did it directly measure 1,000 A flowing while 1,000 V was applied across the device. What makes the result significant is that, rather than improving gallium oxide's thermal conductivity itself, the researchers controlled the distribution of heat and electric field during brief current flow, enabling high-current operation that had been difficult until now.
Junction cooling and a high-permittivity interface cut thermal resistance by more than half
If gallium oxide's thermal conductivity is low, the junction, where the heat is generated, should be placed as close as possible to the cooling substrate. "Flip-chip mounting" puts this idea into practice. The chip is flipped so that its heat-generating side sits close to a copper bonding plate and is attached to a direct bonded copper (DBC) substrate by silver sintering, shortening the distance heat must travel through the thick gallium oxide substrate.
In high-voltage power semiconductors, however, bringing the junction closer to the cooling substrate raises the risk of dielectric breakdown. Even if the metal layer or silver sintered layer bonding the chip protrudes only slightly past the device edge, the electric field concentrates at that metal edge and greatly reduces the device's voltage tolerance. Conventional packages suppress this field concentration by inserting columnar electrodes called metal posts to create distance between the device and the substrate. But the posts lengthen the path heat must take and increase thermal resistance. In other words, improving cooling lowered voltage tolerance, while securing voltage tolerance degraded cooling.
To resolve this trade-off, the team introduced a high-permittivity insulating film at the interface between the gallium oxide device and the package. They used barium titanate () with a thickness of 300 nm. Barium titanate has a relative permittivity above 155 and polarizes strongly in response to an external electric field. The team formed the barium titanate on top of a junction termination extension (JTE) structure made of graded p-type nickel oxide (p-NiO) around the chip periphery, and placed the silver sintered layer over it.
Design analysis shows that the electric field concentrated at the metal electrode edge is dispersed by the polarization of the barium titanate, which eases local field peaks. This made it possible to remove the metal posts, once needed to secure voltage tolerance, from the heat conduction path and to place the silver sintered layer directly near the junction. Mechanical stress caused by differences in thermal expansion is absorbed by a structure combining molybdenum-copper (Mo-Cu) posts with silver ribbon wiring.
The cooling improvement from this design change was confirmed by measurement. The steady-state junction-to-case thermal resistance (), evaluated by transient thermal measurement, was 1.05 K/W for a conventional submodule with metal posts and 0.47 K/W for a submodule with the barium titanate interface. Thermal resistance was cut to less than half while high voltage tolerance was maintained. Integrating a field-relaxing insulating structure with the package's thermal design is a key technology behind the result.
Double-pulse tests demonstrate 1,000 A high-speed switching
The circuit used in the experiments was not an inverter made solely of gallium oxide devices. The team built a "double-pulse test circuit" that uses a commercial SiC power MOSFET as the drive switch and a pre-energized inductor as the current source. The gallium oxide device was incorporated into this circuit as a freewheeling diode, and short-duration high-current conduction and switching behavior were evaluated.
Tests were first conducted on a single submodule using a p-NiO/n- heterojunction diode. At a bus voltage of 1,000 V and a pulse width of 5 µs, it was confirmed that the submodule could carry a stable current of 234 A. For the junction temperature at that point, the paper reports estimates from transient thermal simulation rather than direct measurements. The calculated peak junction temperature reached 251 °C at 234 A and 297 °C at 257 A. This suggests that with extremely short pulses, operation may be possible even when the junction temperature exceeds 250 °C, though these temperatures are not measured values.
The team then fabricated a multi-chip module in which six submodules were arranged symmetrically and connected in parallel. Symmetric wiring to each device suppresses variation in parasitic inductance and keeps current from being biased toward any particular device. Using this module, the team demonstrated short-pulse switching at 1,000 V and 1,000 A at a repetition frequency of 1 kHz.
The turn-off time, determined from the current interruption waveform, was just 23 ns. Almost no reverse recovery current was seen; only a small transient current, mainly due to junction capacitance, was observed. Infrared thermography of the module during repetitive pulse operation showed no large imbalance in heating among the six devices, confirming that current was shared almost equally. During steady repetitive pulse operation, case temperatures stabilized after reaching 92 °C on the top surface and 81 °C on the bottom surface of each submodule.
However, average converted power, conversion efficiency, and energy per pulse cannot be derived directly from these results. The product of 1,000 V and 1,000 A is merely an indicator of power capacity based on switching voltage and conducted current. It does not mean that large currents can be handled continuously, as in the converters used in power grids.
データを表で見る
| Critical peak surge current (A) | |
|---|---|
| 3.6 microseconds | 856 |
| 10 milliseconds | 65 |
Indeed, as the graph above shows, the current a gallium oxide device can withstand varies greatly with pulse width. Even if it can handle hundreds of amperes in pulses of a few microseconds, the limit current drops sharply once conduction time extends to the millisecond range. This is an important result showing that gallium oxide suits short-duration high-current control but still faces constraints in prolonged conduction.
Heat accumulation becomes the problem as pulses get longer
One reason gallium oxide performs well in short-pulse operation is its large volumetric heat capacity, an indicator of how much heat a material can store per unit volume. Although gallium oxide has low thermal conductivity, it is no worse than silicon or SiC at temporarily storing heat. Over a few microseconds, the heat generated does not diffuse far out of the crystal, and most of it is absorbed by gallium oxide's own heat capacity. As a result, the effect of low thermal conductivity is relatively small in short-pulse operation.
As conduction time grows longer, however, a temperature rise from heat accumulation becomes unavoidable. In surge current tests on a single submodule, the team recorded a peak current of up to 856 A at a pulse width of 3.6 µs. But when reverse blocking characteristics were re-evaluated after the final 856 A test, increased leakage current and a clear drop in voltage tolerance were confirmed. In other words, 856 A is the maximum value reached in testing, not a current the device can repeatedly withstand without degradation.
When the pulse width was extended to 10 ms, a duration also used in surge capability evaluations of general power equipment, the critical peak surge current fell to 65 A, about one-thirteenth of the 3.6 µs value. Even if a device can endure large currents in short pulses, heat builds up as conduction time lengthens, and the device's withstand capability drops sharply.
The paper also compares the device with a commercial 1,200 V SiC diode having nearly the same differential on-resistance. Under microsecond-scale short pulses and transient operation at high temperature, conditions were identified in which the gallium oxide submodule maintained higher current density than the SiC diode. The team reports that an optimized module achieved a "pulse power capacity density" above 1.8 MW/cm².
That said, it cannot be concluded that gallium oxide is superior to SiC across the board. The advantage was shown only under specific conditions such as short-pulse conduction and high-temperature operation. Under long continuous conduction or operating conditions where heat accumulates easily, SiC, with its higher thermal conductivity, has the cooling advantage. To make the most of gallium oxide's properties, conduction time, repetition frequency, and cooling conditions must be set appropriately for each application.
Raising the pulse repetition frequency or duty ratio (the proportion of total operating time spent conducting) also leaves the device too little time to cool before the next pulse. Then not only the instantaneous temperature rise from a single pulse but also the average junction temperature goes up. The paper likewise points out that improving external cooling mechanisms is important for drawing out more of gallium oxide's power capacity. Relying on the material's volumetric heat capacity alone limits the ability to handle higher-density pulse operation.
The reliability evaluation of the prototype modules also confirmed the benefit of the high-permittivity interface. In a 10,000-cycle power cycling test, submodules with the barium titanate interface showed little change in forward voltage and kept stable characteristics. In contrast, a comparison module with a silicon dioxide () interface showed marked degradation in forward voltage at about 1,200 cycles. However, this 10,000-cycle test evaluated durability at the prototype stage and does not guarantee continuous operation over tens of thousands of hours or the product lifetimes required in industrial use.
The main experimental results and the extent of what each measurement can establish are summarized below. In particular, measured electrical values must be distinguished from estimates based on thermal simulation. Results obtained in short-pulse operation also cannot be treated as performance in continuous operation.
| Subject | Test conditions / drive mode | Main result | Type of evidence | What this result alone cannot confirm |
|---|---|---|---|---|
| Single submodule | Bus voltage 1,000 V, pulse width 5 µs | Conducted current 234 A, estimated peak junction temperature 251 °C | Measured electrical waveforms, thermal simulation | Directly measured junction temperature, continuous current capability |
| Six-device parallel module | Bus voltage 1,000 V, interrupted current 1,000 A, repetition frequency 1 kHz | Turn-off time 23 ns, top case temperature 92 °C, bottom 81 °C | Measured electrical waveforms and infrared thermography | Continuous average power output, power conversion using only gallium oxide devices |
| Interface material comparison | Steady-state junction-to-case thermal resistance () | Metal post structure 1.05 K/W, structure 0.47 K/W | Measured by transient thermal method | Degradation of the dielectric film in long-term high-temperature operation |
| Surge current limit | Comparison of 3.6 µs and 10 ms pulse widths | 856 A at 3.6 µs, 65 A at 10 ms | Measured by surge current tests | High-current operation in the millisecond range, full integrity after maximum surge |
| Power cycling endurance | Intermittent current stress applied to prototype modules | structure shows little change in characteristics after 10,000 cycles | Durability test of prototype modules | Product lifetime under commercial standards, aging under sustained high electric fields |
From short-pulse demonstration to practical use: remaining challenges
What this research shows is that even gallium oxide, with its low thermal conductivity, can be built into a module with megawatt-class switching power capacity if the device and package are designed together. Rather than improving the material's own thermal conductivity, the approach lowers thermal resistance while suppressing field concentration, and takes advantage of gallium oxide's high volumetric heat capacity and heat resistance. It demonstrates the potential of ultra-wide-bandgap semiconductors, which had mainly been evaluated as material or single-device performance, in actual circuit operation with multiple devices.
The team cites future applications including surge protection for next-generation power grids, medical equipment using pulsed electromagnetic fields, high-voltage pulse power supplies for particle accelerators, and plasma heating power supplies for fusion experiments. Because the ability to handle large power in short bursts is important in these uses, the demonstrated operating characteristics may prove valuable. However, these applications are, for now, future prospects put forward by the research team. What was confirmed in the lab is the operation of a gallium oxide diode in a double-pulse test circuit using a SiC MOSFET; installation in actual industrial equipment has not been demonstrated.
One challenge for practical use is the quality of gallium oxide crystals and higher voltage tolerance. This requires a stable supply of large-diameter, low-defect, high-quality gallium oxide wafers, along with techniques for forming thick, low-concentration epitaxial layers that can block higher voltages. For power systems handling several kilovolts or more, it will also be important to suppress local dielectric breakdown caused by crystal defects and to manage variation in manufacturing processes and device characteristics.
Another challenge is long-term reliability in high-temperature, high-field environments. In this study, junction temperatures above 250 °C were estimated by simulation, and submodules with the barium titanate interface were reported to show stable characteristics in 10,000-cycle power cycling tests. But how much the interface between the semiconductor and the insulating film degrades in environments where strong electric fields and temperature changes repeat over long periods needs further verification. In particular, whether the barium titanate thin film used to relax field concentration can maintain its insulating performance under repeatedly applied high fields will be a key evaluation item determining product lifetime.
Developing active switching devices using gallium oxide is also essential. What was demonstrated is a diode that passes current in one direction; the role of actively turning the circuit on and off was played by the SiC MOSFET. To evolve this into a power conversion system built around gallium oxide, the same thermal management and field relaxation design must be applied to transistors such as MOSFETs and MESFETs. The performance and reliability of whole modules combining diodes and active devices would then need to be verified.
This result is significant in that it linked gallium oxide's high voltage tolerance to megawatt-class switching power capacity in an actual pulsed power circuit. At the same time, that performance depends heavily on short-pulse operating conditions, and the constraints on continuous conduction and prolonged high-current operation have not been removed. Going forward, advancing cooling technology and long-term reliability verification while identifying the short-pulse applications where gallium oxide's characteristics are especially advantageous will be an important step toward practical use.