A breakthrough for the bidirectional switch
By exceeding a blocking voltage of 5 kV while keeping conduction loss and reliability in check, a new generation of monolithic GaN bidirectional switches promise efficient, more compact medium-voltage AC power conversion.
BY YUAN QIN FROM UNIVERSITY OF SCIENCE AND TECHNOLOGY OF
CHINA, MING XIAO FROM XIDIAN UNIVERSITY, KAI CHENG FROM ENKRIS SEMICONDUCTOR,
AND YUHAO ZHANG FROM THE UNIVERSITY OF HONG KONG
With minimal fanfare, power electronics is becoming the enabling layer for a number of converging transitions: electrified transport and industry, the build-out of renewable generation and storage, and the expansion of energy-intensive digital infrastructure. For all these cases, higher efficiency and a higher power density are no longer optional – instead, they directly translate into smaller passive components, simplified thermal management and a lower lifetime cost. The introduction of wide-bandgap semiconductors has already shifted the landscape, with GaN’s success story told through unidirectional HEMTs, powering compact, high-frequency converters from 30 V to 400 V.
But many of the next frontiers for GaN are inherently AC-facing, so switches will have to handle current reversal and bipolar off-state voltages. While it is workable to implement that bidirectional capability with anti-series devices, that solution scales poorly as voltage rises: chip area grows, parasitic capacitances stack, and gate-drive and layout complexity multiply.
A better approach is to attack this problem at its root, using a monolithic bidirectional switch. This form of switch shares the high-voltage region inside one device, and preserves efficiency while simplifying the circuit.
Taking the performance of these switches to a new realm, by delivering a hike in the voltage-handling capability, is our team from University of Science and Technology of China, Xidian University, Enkris Semiconductor and the University of Hong Kong. Our breakthrough promises to open the door to the deployment of this device in medium-voltage AC applications.
Motivation for our efforts has come from considering three questions: Why does bidirectional switching matter in real converters? Why do monolithic implementations offer such a strong advantage? And how is it possible to reach a high voltage with a bidirectional switch?
Why bidirectional?
The applications most in need of a bidirectional switch are those that involve a converter having to connect and disconnect an AC node efficiently. Matrix converters and current-source inverters are classic examples. In these circuits, the switch must handle current in both directions, as the load current and the instantaneous line voltage are not necessarily aligned. In addition, the same device must often sustain positive and negative blocking voltages, as switching states change over a line cycle.
Bidirectional switches are also valuable in solid-state transformers. They are a critical component in Nvidia’s next-generation 800 V power delivery architecture in AI data centres, as well as in medium-voltage motor drives, and emerging AC distribution and protection concepts, where fast, controllable bidirectional elements can replace bulky passive components. For all these examples there tends to be a common economic driver: if a bidirectional function can be implemented with fewer devices, the overall system can be smaller, faster and more efficient.
Figure 1. Conceptual comparison between a bidirectional switch realised using two discrete GaN transistors in anti-series and a single monolithic bidirectional switch sharing one drift region.
At medium voltages, strong opportunity for the deployment of the bidirectional switch is in the AC-DC front end, used in data centres, rail traction and industrial power supplies. Here, design engineers are using three-phase active rectifiers and solid-state transformer stages to increasingly target higher switching frequencies and shrink magnetics.
In these systems, the bidirectional function is often realised with anti-series device pairs, a compromised solution that increases the active device count and stacks parasitic capacitances and stored charge. With this approach there’s a higher commutation energy penalty on every transition, making it harder to push up the switching frequency without sacrificing efficiency.
Additional examples associated with the potential deployment of bidirectional switches are medium-voltage motor drives and variable-speed compressors. In these cases, introducing bidirectional switches enables matrix-converter and current-source topologies that remove the bulky DC-link electrolytic capacitor. Eliminating or downsizing the DC link improves lifetime and power density. However, to ensure success, the bidirectional switch must block bipolar voltages safely and repeatedly under fast switching, elevated temperature and rapidly changing voltages.
The monolithic advantage
When a brute-force anti-series architecture is adopted, this doubles the number of active devices. That’s not an elegant solution – as each device contributes roughly half the total on-resistance, there is a potential quadrupling of chip area to realise the same conduction loss as a single unidirectional transistor. At the system level, the penalty can be even more severe, with extra devices introducing additional capacitance, which must be charged and discharged for every switching cycle.
Introducing monolithic bidirectional switches reduces this overhead by placing two gated channels on one shared drift region – this avoids duplication in the high-voltage region (see Figure 1). At lower voltages, monolithic GaN bidirectional switches are already appearing as engineering samples in industry, pointing to a path for adoption.
Figure 2. Device schematic and equivalent circuit model of multi-channel GaN monolithic bidirectional switch demonstrated at IEDM 2025.
Figure 3. Bidirectional forward and blocking characteristics of the multi-channel GaN monolithic bidirectional switch.
One pertinent question is whether it’s possible to preserve these monolithic advantages when shifting the voltage rating into the kilovolt class. In this regime, electric-field management, buffer breakdown, and reliability could be limiting factors.
Taming fields
In a bidirectional device, the high-voltage stress does not sit at a single drain edge in a fixed direction. Instead, the peak field can move, depending on the polarity of the blocked voltage, and it can couple to multiple gate stacks. Due to this, edge termination and field shaping are central design problems.
When we recently produced a 3.3 kV enhancement-mode monolithic bidirectional switch, we demonstrated one way to tackle this challenge. Working on a mature p-GaN gate platform on sapphire, the dual junction termination extension held the key to our success, with implementation in the p-GaN layer of the standard gate stack, without epitaxial regrowth. Our devices delivered a breakdown voltage exceeding 3.3 kV in both polarities, a threshold voltage around 0.6 V, and a specific on-resistance of 5.6 mΩ cm2, while maintaining symmetric conduction in both directions.
Figure 4. Benchmarking state-of-the-art (ultra-)wide-bandgap monolithic bidirectional devices and the practical performance limits of discrete anti-series implementations.
Looking beyond the headline numbers, the junction termination extension concept is worthy of consideration. This refinement provides a field-shaping knob that does not rely solely on long field plates. Instead, by balancing field peaks at the gate edge and the termination edge, it allows the average lateral field to be pushed beyond what is typical in field-plated lateral GaN devices, while maintaining a low leakage current under bipolar blocking.
Multi-channel leap
Our 3.3 kV device established a high-voltage foundation. But it’s just a starting point, as medium-voltage power conversion often demands higher voltages and higher currents. Ideally, this is realised without comprise, but in lateral GaN that’s not easy, as efforts to reach to higher voltages tend to drive the drift region longer and the specific on-resistance higher, unless there’s a reduction in channel and access resistances. One way to address this is to move to multi-channel AlGaN/GaN materials, because multiple parallel channels can combine a high current capacity with a low sheet resistance.
Late last year, at the International Electron Devices Meeting (IEDM), we unveiled the first multi-channel GaN monolithic bidirectional switch. It’s a five-channel device, fabricated on a 150 mm GaN-on-sapphire wafer grown by Enkris Semiconductor (see Figure 2).
Strengths of this switch include a symmetric blocking of up to 5 kV in both polarities, a specific on-resistance of around 20 mΩ cm2 and a threshold voltage of about 0.8 V (see Figure 3). To our knowledge, 5 kV is the highest breakdown voltage ever reported for a monolithic bidirectional switch (see Figure 4).
Our device architecture combines two important ideas (see Figure 2). First, by employing a dual-junction termination extension, connected to the two high-voltage gates, it provides effective electric-field management in the central multi-channel region. Second, it realises enhancement-mode operation using a monolithic dual-cascode architecture. By integrating two low-voltage, single-channel enhancement-mode HEMTs on either side of the multi-channel bidirectional region, the overall switch behaves as an enhancement-mode device while the high-voltage multi-channel section retains its natural depletion-mode characteristics.
Figure 5. Electric-field management using a p-GaN dual junction termination extension: simulated E-field contours illustrating the impact of the junction-termination extension (JTE) length and the risk of field crowding in the spacing region for an overlong termination.
Optimising the termination length is particularly important in a bidirectional device (see Figure 5). When a high-voltage GaN switch is unidirectional, device designers may turn to a long p-GaN reduced-surface-field region to aid voltage scaling. But that’s not an attractive option in a bidirectional structure with two terminations facing one another, as long termination on the high-side cannot necessarily be depleted in a helpful way – instead, there’s the threat of failing to take advantage of the footprint and crowding the field in the spacing between the two terminations.
Based on simulations that suggest an optimal termination length that balances these effects, we implemented a termination length that supports 5 kV bipolar blocking. We recorded no destructive breakdown under our measurement compliance.
Reliability focus
A high voltage is only half the story. Reliability also matters, requiring serious consideration. In bidirectional devices, the stress differs from that found in a conventional transistor, because the blocked polarity alternates in normal operation and the second gate can sit at a non-trivial potential relative to the substrate. These factors raise practical questions about dynamic on-resistance, threshold stability and the right stress protocols for qualification.
Our multi-channel bidirectional switch shows encouraging signs, regarding reliability. In our paper from IEDM 2025, we described a comparison of our multi-channel 5 kV device and a single-channel bidirectional device with a similar termination concept. Under a high-temperature reverse-bias stress at 150 °C, our multi-channel device underwent a shift in on-resistance of less than a 5 percent after 1,000 s of stress at 1.7 kV. In contrast, the single-channel device exhibited substantially larger on-resistance shifts under comparable stress conditions (see Figure 6).
Equally encouraging for system designers, we have found that our multi-channel device is relatively insensitive to substrate bias over our testing range. It’s known that substrate bias management is a challenge for GaN-on-silicon bidirectional devices in practical circuits, because the substrate and buffer can couple to the switching node. By moving to an insulating sapphire substrate and exploiting a multi-channel stack, our device offers a proven robustness to substrate bias that’s attractive for real converter environments.
To interpret our observed stability, we have proposed a physical picture, in which the multiple channels provide an inherent shielding effect. Our view is that in a single-channel device, charge trapped at the surface or in the buffer directly modulates only the conduction channel, leading to larger dynamic and stress-induced resistance shifts. Meanwhile, in a multi-channel structure, the top channel screens the influence of surface traps and the bottom channel screens the influence of buffer traps, to leave the middle channels less perturbed. The upshot is a smaller change in on-resistance after stress (see Figure 7).
Where next?
Thanks to a combination of 5 kV bipolar blocking, enhancement-mode control, and improved dynamic stability, our monolithic GaN bidirectional switches are compelling candidates for medium-voltage AC power conversion. Here, their established rivals are silicon or SiC solutions.
At the device level, the next steps are clear: a further reduction in specific on-resistance, realised through layout and material optimisation; refinement to termination designs, to scale blocking voltage while maintaining controlled leakage; and a broadening of reliability evaluation, beyond static stress, to include circuit-relevant dynamic and switching tests. In practice, that means characterising dynamic on-resistance, switching energy and gate/termination bias sensitivities under realistic temperature conditions, while considering derivatives of voltage and current with respect to time. In addition, these measurements must be correlated with long-duration stress, to isolate and uncover dominant trapping and degradation mechanisms.
Figure 6. Device reliability: high-temperature reverse-bias stress sequence, and the evolution of on-resistance shift for single-channel versus multi-channel devices.
Figure 7. Schematic illustration of device physics: impacts of interface and buffer traps on middle channels are shielded by top and bottom channels, reducing the dynamic on-resistance shift for the multi-channel GaN monolithic bidirectional switch.
Once that device- and circuit-level understanding is in place, efforts can be directed at the system-level, re-thinking topologies constrained by cost and the parasitics of anti-series device pairs. A genuinely compact monolithic bidirectional switch has the potential to simplify gate driving, reduce the energy stored in parasitic capacitances, and raise switching frequency without an equivalent penalty in loss. If these benefits are realised at scale, they could translate into smaller passive components, a higher power density, and increased efficiency in the medium-voltage converters underpinning electrified transport, renewable integration and industrial automation.
While bidirectional switching is not a new functional requirement in power electronics, it is becoming a more urgent one as systems shift towards higher-frequency operation and tighter volumetric constraints. Given the great progress in moving from 650 V to 5 kV in GaN monolithic bidirectional switches, alongside a growing understanding of their unique reliability physics, there is good reason to think that this technology is moving from an academic curiosity to a platform that will meaningfully reshape medium-voltage converter design.
