I recently came across a piece of news: Zhanxin Electronics officially announced the mass production and shipment of its 3.3kV SiC MOSFET.
To be honest, over the past few years, we have seen too many announcements regarding the mass production of domestic silicon carbide (SiC) devices, ranging from 650V to 1200V and then to 1700V. Every round claims a "breakthrough," and every round fiercely competes for automotive-grade certifications. At first, it was somewhat exciting, but later it just became numbing. After all, it all revolves around electric vehicles (EVs), merely squeezing out a little more performance in the parameters.
However, the figure of 3.3kV still made me pause.
The reason is simple: this voltage class is already beyond the reach of "automotive-grade" applications. No matter how aggressive the 800V platform for EVs gets, it maxes out at 1700V. The 3.3kV class targets an entirely different set of applications—traction converters for rail transit, converter valves for power grids, medium-voltage grid connection for wind, solar, and energy storage, and solid-state transformers (SST).
The common characteristics of these applications are: massive scale, stringent requirements, and decades of dominance by silicon-based IGBTs, with domestic devices having almost no presence.
So the questions arise: Why now? Why 3.3kV? And what does it really mean for a domestic IDM to achieve mass production at this voltage class?
OK, starting from this news today, let's try to break down and clarify these questions.
First Question: What Exactly Makes the 3.3kV Voltage Class So Special?
In the world of power semiconductors, voltage classes are never just a numbers game. Below 1200V, it is the battlefield for consumer electronics and NEVs (New Energy Vehicles), where cost-effectiveness, yield rates, and automotive-grade certifications are fiercely contested. Whoever achieves a few milliohms lower on-resistance can secure a larger share of the supply chain.
But above 1700V, the rules of the game begin to change. The competitors are no longer silicon MOSFETs from the same family, but IGBTs and IGCTs, which have dominated the high-voltage and high-power domain for decades.
IGBTs are not bad. Over the past thirty years, high-speed rail, power grids, and heavy industry have relied entirely on them. However, their physical ceiling is evident: there is a tail current during turn-off, and the switching frequency is stuck around 3kHz to 5kHz. Pushing it any higher would cause losses to explode exponentially.
Therefore, system engineers are forced to accept the reality of 50Hz or a few hundred Hz industrial frequencies, using massive iron-core transformers and thick filter inductors in exchange for safety. The overall weight of the equipment often reaches tons, and the footprint is measured in square meters.
The emergence of 3.3kV SiC MOSFETs is equivalent to introducing a completely new variable into an equation that has been locked by IGBTs for decades.
Its switching speed can reach tens of kHz or even higher, with switching losses only one-tenth to one-thirtieth of those in silicon devices with the same voltage rating. This means isolation transformers can leap from the industrial frequency of 50Hz to the medium-to-high frequency band above 20kHz, reducing volume and weight to one-tenth to one-fifth of the original. Architectures that previously required three or four conversion stages might be accomplished with just one or two.
This is not a linear improvement in parameters, but a fundamental reconstruction of the system architecture.
Therefore, the uniqueness of the 3.3kV threshold lies in the fact that it sits exactly at the intersection where "automotive-grade cannot reach, and IGBTs start to struggle."
Moving up, it leads to truly capital-intensive fields such as rail transit, power grids, and AI data centers; looking down, it faces the 1200V red ocean market that has already been fiercely competed in by SiC. Simply put, whoever can establish a foothold at this voltage class will secure a ticket to the next stage of competition.
And this announced mass production happens to fall exactly on this intersection.
The 3.3kV SiC MOSFET recently mass-produced by Zhanxin Electronics, along with the 10kV-class research results released jointly with Zhejiang University, provide two sets of specific domestic technical parameters: the former includes a TO-247-4 HV discrete device with a 50mΩ on-resistance and an XPak half-bridge module with a 750A current capacity, while the latter achieves a specific on-resistance of less than 120mΩ·cm², approaching the theoretical limit of silicon carbide materials.
Looking at these parameters individually, they are just a set of impressive numbers. But if we evaluate them within actual power electronics systems, a crucial difference becomes apparent.
In the 1200V era, what SiC did was to "replace silicon devices" within existing architectures, reducing losses and improving efficiency, with the system topology basically unchanged; but when the voltage rating reaches 3300V or even 10kV, the nature of the matter changes. It begins to allow system engineers to rethink the entire energy conversion architecture.
The reason is that the biggest cost and volume burdens in medium- and high-voltage systems have never been the switching devices themselves, but the industrial frequency transformers, multi-stage series structures, and massive filter components that had to exist to accommodate the low switching frequencies of silicon devices. When a single switching transistor can withstand 3300V and switch at tens of kHz simultaneously, those compromised designs that existed to cater to IGBTs suddenly become unnecessary.
This is precisely the system logic behind the parameters.
When the voltage rating of the core switching device leaps from 1200V to 3300V or even 10000V, while simultaneously maintaining the inherent high-frequency and low-loss characteristics of silicon carbide, the energy routing mode that originally had to rely on multi-stage device series connection, massive industrial frequency transformer buffering, and complex voltage-sharing circuit protection suddenly gains the right to simplify.
This dimensional upgrade in voltage is essentially using the extreme progress in semiconductor processes to replace redundant copper and iron weight and energy consumption in physical systems.
Second Question: Is There Really a Shortage of 3.3kV SiC MOSFETs in China?
The answer is: not a shortage, but not abundant either.
One thing needs to be clarified first: Zhanxin Electronics is not the first to mass-produce 3.3kV SiC MOSFETs in China. Prior to this, CRRC Times Electric had already applied products of the same class to rail transit traction systems and achieved batch delivery. Austot announced the pioneering mass production of 3300V/60A SiC MOSFETs in 2024, and SICC's product line has also covered the full 3300V series. In addition, peers such as SILAN have also launched 3300V SiC MOSFETs in the first half of 2026 and entered the sampling and verification stage. Therefore, a supply pattern of multiple manufacturers coexisting has been formed in China at this voltage class.
*Layout of domestic SiC manufacturers in 3.3kV MOS, compiled based on public information
So what is the significance of this announcement?
In my opinion, there are two points worth highlighting separately.
First, it is done using the IDM (Integrated Device Manufacturing) model. Chip design, wafer fabrication, and module packaging are all in-house. This is a significant bonus in the high-voltage SiC field.
The reason is simple: the process window for high-voltage devices is extremely narrow. Any deviation in epitaxial thickness, implantation energy, or terminal structure will directly reflect on yield and reliability. Under the foundry model, the communication cost between design and manufacturing is high, and it is difficult to quickly locate problems when they occur. Under the IDM model, the speed of process iteration and the efficiency of problem closed-loop are both significantly higher. For scenarios like rail transit and power grids that have extremely high requirements for long-term reliability, whether the supplier has its own production line is an important screening criterion.
Second, it provides a complete "device + driver" solution. The 3.3kV XPak half-bridge module achieves a creepage distance of 15.6mm, meeting high-voltage insulation safety regulations; the supporting IVCO1B12 driver chip integrates gate oxide leakage current monitoring functionality.
Downstream customers receive not loose components, but a matched and tested subsystem. This means system integrators do not need to spend a lot of time debugging driver parameters and designing protection circuits themselves, lowering the entry barrier. For scenarios with extremely low fault tolerance like SST and rail transit, this "out-of-the-box" experience is often an important consideration for customers willing to place their first order.
At the PCIM Europe exhibition in June 2026, Zhanxin used this solution to build a megawatt-class rectifier-inverter prototype, also releasing a very strong signal: domestic high-voltage SiC is no longer just benchmarking international first-tier products on parameter sheets, but is starting to win commercial orders in the most tedious yet most critical aspects such as system-level reliability, safety regulation adaptation, and long-term aging.
Therefore, Zhanxin's position this time is very clear: it is not the first to take the plunge, but it provides an additional option with full-chain capabilities within the existing domestic 3.3kV supply chain, and has achieved differentiation in packaging and driver support.
Third Question: After 3.3kV, Why Look at 10kV?
This time, Zhanxin not only announced the mass production of 3.3kV, but the press release once again mentioned the technical achievements of the 10kV SiC MOSFET jointly developed with Zhejiang University, achieving a specific on-resistance of less than 120mΩ·cm², approaching the theoretical limit of the material.
Many people's first reaction to seeing this number is: showing off. After all, 10kV is still a distance away from large-scale mass production. The cost of thick epitaxy, the yield of ultra-large chip areas, and the creepage and insulation challenges in packaging—each is a mountain difficult to climb in the short term.
But if we extend the timeline to five to ten years, the significance of 10kV is different.
Current AI data centers are facing a very awkward power distribution dilemma. The power grid delivers 10kV or 13.8kV medium-voltage AC, while the GPU chips in the racks consume 48V or 800V DC. How long is the intermediate conversion chain? Industrial frequency step-down transformers, low-voltage distribution cabinets, UPS, rectifier modules—after layer-by-layer losses, not to mention efficiency loss, the footprint and heat dissipation are huge hidden costs.
NVIDIA has repeatedly mentioned a concept called "medium-voltage direct-connected rectification" in its 800V HVDC white paper. Simply put, it wants to eliminate those redundant intermediate conversion stages and rectify directly from 10kV to 800V DC.
*800VDC Power Distribution Architecture in NVIDIA's Next-Generation AI Data Centers
But the physical premise of this concept is: a switching device that can withstand 10kV and operate at high frequencies.
If 1200V SiC is used for a cascaded H-bridge, it requires seven or eight, or even ten, stages of series-connected units. The voltage-sharing circuit is so complex that it makes one's scalp tingle, the control bandwidth is largely consumed by the voltage-sharing capacitors, and the failure rate multiplies. If IGBTs are forced to handle it, the frequency cannot go up, and the concept of SSTs—"high-frequency isolation replacing industrial frequency iron cores"—simply cannot hold water.
If 10kV SiC MOSFETs mature, theoretically, a small number of devices can be used to build a two-level or three-level medium-voltage direct-connected topology. The number of cascaded stages is reduced by 60%, the power units and peripheral driver PCBs are significantly simplified, and the system BOM cost is expected to drop by 20% to 35%. This is the true physical cornerstone for the future implementation of the 10kV direct-to-800V DC power supply architecture.
Therefore, 3.3kV is the present continuous, solving the urgent needs of current rail transit, SST, and medium-voltage grid connection for wind, solar, and energy storage; 10kV is the future tense, securing a ticket to the next generation of smart grids and AI high-density power supply.
Zhanxin holds both in its hands, and the intention is obvious: it does not want to be just a supplier for a certain generation, but wants to establish long-term process platform depth in the high-voltage SiC track.
Fourth Question: Is This Path Easy to Walk?
It is not easy.
If we cut open the path between 3.3kV and 10kV, what lies in the middle is far more than just a multiple of voltage numbers, but a triple cliff of materials, processes, and packaging.
Let's talk about materials first. 10kV SiC MOSFETs require ultra-thick epitaxial layers of over 100 micrometers, while medium- and low-voltage devices only need 10 to 20 micrometers. The growth cycle for thick epitaxy is extended to over ten hours, and the micropipe dislocation defect density must be pressed below 0.1 per square centimeter; otherwise, electric field concentration under high voltage will directly cause breakdown. For material yield at this level, the number of substrate manufacturers globally that can supply stably can be counted on one hand.
Then there is the process. The higher the voltage rating, the larger the area occupied by the terminal structure at the edge of the chip, and the proportion of the active area is diluted. The contradiction between specific on-resistance and breakdown voltage needs to be balanced using fine work such as narrow JFET regions, field plate optimization, and high-energy implantation. The 120mΩ·cm² achieved by Zhanxin and Zhejiang University is already close to the theoretical limit, but every step closer is exchanged for yield.
In other words, being able to make it and being able to mass-produce it at low cost are two different things.
Finally, there is packaging. At 10kV, the creepage distance must be over 20 millimeters. Traditional TO-247 and aluminum oxide DBC substrates simply cannot handle it. Aluminum nitride or sapphire substrates, silver sintering die attach, and bondless planar interconnects must be used to push stray inductance down to a few nanohenries. Otherwise, the dv/dt of high-speed switching rushing to 100kV/μs, along with voltage overshoot and common-mode interference, can directly smash the driver and protection circuits to pieces.
These hidden barriers are the fundamental reason why high-voltage SiC has not been widely deployed on a large scale over the past decade or so. It is not that no one wants to do it, but the manufacturing and material foundation of the entire industry chain cannot support it. I have discussed in detail the 10kV SiC device technology, mass production bottlenecks, and application trends in my previous articles. For those interested, you can click on "Domestic 10kV SiC: How Far from Sample to Mass Production?"
*This 10kV SiC MOSFET is produced based on Zhanxin Electronics' third-generation planar gate process platform. The single chip size reaches 10mm×10mm, the on-state current is close to 40A, and the breakdown voltage exceeds 12kV. It is currently the largest published 10kV-class SiC MOSFET chip, with a specific on-resistance of less than 120mΩ·cm², approaching the theoretical limit of SiC materials. Image source: Zhejiang University PEDL Team, Zhanxin Electronics.
But what is interesting is that the collaboration model of the industry chain is also changing this time.
In the past, making high-voltage devices was basically a matter of device manufacturers working in isolation, and then going to system manufacturers to sell them after development. It is different now. System integrators represented by NARI Technology and Jinpan Technology have begun to carry out collaborative definitions with upstream device manufacturers in industry consortiums such as solid-state transformers. Starting from application requirements, they reversely drive the formulation of material and device specifications.
In this way, substrate manufacturers know what kind of thick epitaxy will be needed in the future, device manufacturers know which reliability indicators the system side cares about most, and OEMs know how to set the voltage and power ratings for next-generation products.
This closed-loop linkage between upstream and downstream is shortening the conversion cycle of high-voltage SiC from laboratory samples to actual engineering applications.
Final Question: So, What Stage Are We at Now?
Objectively speaking, although today's 3.3kV SiC MOSFETs have already secured commercial orders, high-reliability indicators such as gate oxide lifespan under long-term high-frequency switching, cosmic ray failure rates, and extreme short-circuit tolerance still require tens of thousands of hours of field operation data to be validated. This cannot be completed in just a year or two.
The 10kV class must also cross the mountain of cost and yield. Currently, thick epitaxy and ultra-large chip areas lead to extremely high per-wafer costs. For a considerable period, it will coexist with 3.3kV and 1700V solutions, first being used in cutting-edge scenarios with the highest value density and the most sensitivity to volumetric efficiency, such as aerospace power supplies, high-end SST demonstrations, and pulsed power, before gradually sinking downwards.
But the direction is becoming increasingly clear.
Energy routing in power systems is evolving from "passive industrial-frequency copper and iron lumps" to "programmable solid-state energy routers."
And high-voltage SiC is the pacemaker for this router.
When domestic silicon carbide manufacturers in wafer fabs expand the 3.3kV process window bit by bit and push the 10kV specific on-resistance closer to the theoretical limit, they are actually helping to rewrite a physics textbook for the entire new power system and AI computing infrastructure.
In that textbook, transformers no longer have to be as bulky as oxen, power grids no longer have to be low-frequency and slow, and the flow of energy can finally run on high-frequency, highly efficient, and controllable semiconductor switches.
Returning to the news about Zhanxin. The mass production and shipment of 3.3kV may just be the beginning.
But it marks a coordinate: domestic high-voltage SiC has moved from "can make it" to "is using it."
The next chapter of the story depends on whether orders can turn into mass volumes, whether 10kV can move from papers to production lines, and whether this entire industry chain closed-loop composed of materials, processes, packaging, and systems can truly run through.
The road is still long, but the direction is right.
Sources:
[1] Zhanxin Electronics
[2] Zhejiang University
[3] NVIDIA