Wide-Bandgap Power Electronics in 2026: How SiC and GaN Are Redrawing the Power Conversion Map

Every electric vehicle, solar inverter, server rack and industrial drive contains a silent tax: the energy lost each time electricity is converted from one form to another. For roughly six decades, silicon set the ceiling on how small that tax could be. That ceiling is now being lifted — not by a single breakthrough, but by the slow industrialisation of two materials, silicon carbide (SiC) and gallium nitride (GaN), whose atoms are bonded tightly enough to withstand electric fields an order of magnitude stronger than silicon can survive.

The engineering case for wide-bandgap (WBG) semiconductors has been settled in the laboratory for two decades. What has changed since roughly 2023 is the commercial case, and it has changed in both directions at once. SiC is now standard in high-voltage electric vehicle powertrains, yet the companies making it have endured price collapse, overcapacity and, in one prominent case, bankruptcy. GaN is winning designs in AI data centre power delivery while its leading suppliers fight each other in courts on three continents.

This article is written for practising engineers, technical managers, procurement leads and investors who need an accurate picture rather than a promotional one. It covers the device physics that determines what these materials can and cannot do, the design changes they force on the rest of the system, the standards that now govern their qualification, the economics that will determine adoption pace, and an honest account of what remains uncertain. Recent developments are dated precisely; where evidence is thin or contested, that is said plainly.

---

Executive Summary

• Wide-bandgap devices are no longer emerging technology in their core sockets. SiC is established in high-voltage EV traction inverters, and GaN is established in consumer fast chargers; both are now competing for AI data centre power delivery.
• The physics advantage is real and bounded. SiC and GaN reduce switching and conduction losses and permit higher-frequency operation, which shrinks magnetics and heatsinks. They do not eliminate loss, and they introduce new failure modes and design constraints.
• The industry's financial health has diverged sharply from its technical progress. Wolfspeed, a foundational SiC supplier, filed for Chapter 11 on 30 June 2025 and emerged on 29 September 2025 after cutting debt by roughly 70%.
• Chinese suppliers have reshaped competitive dynamics in both SiC substrates and GaN devices, contributing to steep price declines and, in June 2026, a Chinese Supreme People's Court ruling that upheld a sales injunction against certain Infineon GaN products.
• The 800 VDC data centre transition is the most consequential new demand driver. Moving in-rack distribution from 54 V to 800 V makes wide-bandgap devices effectively mandatory rather than optional.
• Published market forecasts diverge by more than a factor of two, partly because "silicon carbide market" figures sometimes include abrasives and refractories alongside power semiconductors. Treat any single number with caution.
• Qualification standards have matured substantially. JEDEC's JC-70 committee has published a body of SiC and GaN test guidelines, complementing the automotive AQG 324 qualification framework.
• Ultra-wide-bandgap materials such as β-Ga₂O₃ and diamond are scientifically promising but pre-commercial, constrained by thermal conductivity, p-type doping and manufacturing maturity.
• The dominant engineering challenge has shifted from the device to everything around it: gate drive, layout parasitics, EMI, packaging and thermal management now determine whether the theoretical gain is realised.

---

Why Bandgap Matters: The Technical Foundation

A semiconductor's bandgap is the energy required to lift an electron from its bound valence state into the conduction band where it can carry current. Silicon's bandgap is about 1.12 eV. In 4H-silicon carbide it is roughly 3.26 eV, and in gallium nitride roughly 3.4 eV — hence "wide bandgap."

The bandgap itself is not the useful property. What follows from it is.

Critical electric field. A wider bandgap means the material tolerates a far stronger electric field before avalanche breakdown — roughly ten times silicon's. This is the single most important consequence. To block a given voltage, a power device needs a drift region thick enough and lightly doped enough to sustain the field. If the material tolerates ten times the field, the drift region can be roughly ten times thinner and far more heavily doped. Resistance falls dramatically. This relationship is captured by Baliga's figure of merit, which scales with the cube of the critical field.

Thermal conductivity. 4H-SiC conducts heat roughly three times better than silicon, which matters because power devices fail thermally long before they fail electrically. GaN's advantage here is smaller and depends heavily on the substrate beneath it — most commercial power GaN is grown on silicon, so the silicon substrate governs heat extraction.

Device architecture. This is where SiC and GaN diverge in ways that determine their application boundaries. SiC power devices are vertical: current flows from a top-side source through the bulk crystal to a bottom-side drain, so blocking voltage scales with wafer thickness. Commercial GaN power devices are lateral high-electron-mobility transistors (HEMTs), in which a two-dimensional electron gas conducts across the surface of the chip. Lateral geometry gives GaN exceptionally low parasitic capacitance and superb switching speed, but scaling blocking voltage means consuming chip area, which is why commercial power GaN clusters at 650 V and below while SiC comfortably serves 1,200 V and above.

Table 1 — Material and device properties: an engineer's comparison

Table constructed for OneWise from consensus values in device-physics literature. Figures are representative rather than device-specific; consult vendor datasheets for design work.

Editorial note on figures of merit. Baliga's figure of merit ranks materials by theoretical conduction-loss potential at a given blocking voltage. It says nothing about switching loss, packaging parasitics, gate reliability, thermal extraction or cost — all of which dominate real system performance. A material can lead on FOM by an order of magnitude and still lose in production, which is precisely the situation of gallium oxide today.

---

From Laboratory Curiosity to Volume Production

The historical arc matters because it explains why adoption feels simultaneously fast and slow.

SiC's obstacle was never physics; it was crystal growth. Growing a low-defect SiC boule requires sublimation at temperatures above 2,000 °C, and micropipe defects made early wafers unusable for high-voltage devices. Commercial SiC Schottky diodes appeared in the early 2000s; SiC MOSFETs followed roughly a decade later, once gate-oxide reliability at the SiC/SiO₂ interface became tractable. The interface remains SiC's most-studied reliability concern, and it is the reason JEDEC devoted separate publications to threshold-voltage measurement and gate-oxide evaluation.

GaN's path ran through optoelectronics. The blue LED work that earned the 2014 Nobel Prize in Physics established GaN epitaxy at industrial scale. Power GaN then borrowed that infrastructure — critically, by growing GaN on inexpensive large-diameter silicon substrates rather than on native GaN. Early GaN HEMTs were depletion-mode (normally on), which complicated circuit design; the shift to enhancement-mode, normally-off devices removed a significant adoption barrier.

The commercial inflection for SiC came with high-voltage EVs. Once automakers moved from 400 V to 800 V battery architectures — for faster charging and thinner cabling — silicon IGBTs became a poor fit and SiC MOSFETs became the natural choice. That single design decision, replicated across dozens of vehicle platforms, created the demand that justified building 200 mm SiC fabs.

---

Where Each Material Wins Today

The most common analytical error in this field is treating SiC and GaN as direct substitutes. They overlap at the edges but occupy different design spaces.

Table 2 — Application fit by socket

Original OneWise synthesis. Boundaries are moving; the table reflects prevailing practice as of mid-2026, not a permanent division.

The pattern is consistent: SiC wins where blocking voltage and current density dominate; GaN wins where switching frequency and power density dominate. As of mid-2026, no automaker has announced GaN in a production traction inverter, while GaN's presence in high-density DC/DC conversion is expanding rapidly.

---

The 800-Volt Data Centre: The Decisive New Socket

For a decade, the marginal buyer of wide-bandgap devices was the automotive industry. That is changing.

AI compute racks have outgrown the 54 VDC in-rack distribution standard. At megawatt-scale racks, 54 V distribution demands impractical quantities of copper busbar and consumes rack space that would otherwise hold compute. NVIDIA has published an 800 VDC architecture intended to address this, with an ecosystem of silicon, component and power-system partners including Analog Devices, Infineon, Innoscience, Monolithic Power Systems, Navitas, onsemi, Renesas, ROHM, STMicroelectronics and Texas Instruments. Texas Instruments announced a complete 800 VDC reference architecture on 16 March 2026, timed to NVIDIA's GTC conference.

Two technical consequences follow, and they are the reason this transition matters beyond hyperscale operators:

1. Silicon is largely excluded. At 800 V with high switching frequency, SiC and GaN are not an optimisation — they are the enabling technology.
2. Conversion stages are being eliminated. Vendors have demonstrated direct 800 V-to-low-voltage conversion that removes the intermediate 48 V bus converter entirely. Navitas showed an 800 V-to-6 V power delivery board at GTC on 16 March 2026, using sixteen 650 V GaN FETs in a stacked full-bridge with silicon MOSFETs on the secondary side, targeting around 96.5% peak efficiency at a 1 MHz switching frequency. STMicroelectronics has shown 800 V boards combining 1,200 V SiC for hot-swap protection with 650 V GaN in a stacked half-bridge.

Read vendor efficiency claims carefully. Figures such as "97.5% peak efficiency" or "2,500 W/in³" are typically peak-load, board-level laboratory measurements under favourable conditions. End-to-end system efficiency across a realistic load profile is a different and usually lower number. NVIDIA's own framing — up to 5% end-to-end efficiency gain and up to 30% lower total cost of ownership — is a supplier projection for a system architecture not yet in volume production, not a measured field result.

Full-scale production is generally expected around 2027 with NVIDIA's next rack generation, with supplier ramp underway through 2026. That timeline is a plan, not a fact, and it depends on liquid cooling, connector and protection ecosystems maturing in parallel.

Figure 1 — Power conversion chain: 54 V versus 800 VDC rack architecture Purpose: Show why higher distribution voltage removes conversion stages and reduces copper. Layout: Two horizontal flow bands, stacked. Upper band (legacy, labelled "54 VDC distribution"): Grid AC → Transformer → UPS → PDU (415/480 VAC) → Rack power shelf (AC/DC) → 54 V busbar (drawn as a thick copper-coloured bar) → Intermediate bus converter (54 V → 12 V) → Voltage regulator module → GPU (0.8 V). Lower band (labelled "800 VDC distribution"): Grid AC → Solid-state transformer / rectifier → 800 VDC busbar (drawn visibly thinner) → Direct 800 V → 6 V power delivery board (annotate "GaN") → Voltage regulator module → GPU (0.8 V). Annotations: Mark eliminated stages in the lower band with a dashed grey outline and the label "stage removed." Place a small semiconductor icon beside each conversion block, colour-coded: grey = silicon, blue = SiC, green = GaN. Add a busbar cross-section callout comparing copper mass. Caption: "Raising in-rack distribution voltage removes conversion stages and reduces conductor mass — but only wide-bandgap devices switch efficiently at 800 V."

---

The Economics: Technical Success, Financial Stress

A striking feature of this market is that technical adoption and supplier profitability have moved in opposite directions.

Wolfspeed, one of the industry's foundational SiC companies, filed for Chapter 11 protection on 30 June 2025 and completed its restructuring on 29 September 2025, reducing total debt by approximately 70%, extending maturities to 2030 and cutting annual cash interest expense by roughly 60%. The company cited production delays, intensified Chinese competition and softer EV demand among the pressures leading to the filing. It emerged intact, retaining its vertically integrated 200 mm manufacturing footprint.

Navitas, a pure-play wide-bandgap supplier, saw revenue fall from $83.3 million in 2024 to $45.9 million in 2025 as it deliberately retreated from the commoditised mobile-charger market toward data centre and industrial power — a strategic repositioning with a real near-term cost.

Capacity utilisation tells the underlying story. Yole Group measured upstream SiC substrate and epitaxy utilisation at roughly 50% in 2025, with device lines near 70%. Building capacity ahead of demand is normal in semiconductors; sustaining it through a demand pause is what separates survivors from casualties.

On market-size figures: a caution

Published forecasts for "the silicon carbide market" in 2026 range from under $3 billion to over $5.7 billion, with projected growth rates from under 8% to over 40%. Much of this spread is definitional rather than analytical. Broad SiC market reports often include abrasive and refractory silicon carbide — an old, large, slow-growing industrial commodity — alongside power semiconductors. Others count only devices; others include substrates and modules.

Practical guidance: before citing a market figure, establish whether it covers power devices only, the full SiC value chain, or all industrial SiC. Yole Group's forecast of an approximately $11 billion power SiC device market by 2031 is a device-scoped estimate and should not be compared directly with headline figures built on a broader definition.

Cost trajectory

The cost gap with silicon is narrowing but has not closed. Analyst estimates put SiC devices at roughly three to five times the cost of equivalent silicon parts, with the transition from 150 mm to 200 mm wafers expected to deliver a 30–40% per-die cost reduction as yields mature. These are projections built on assumed yield curves, not observed outcomes, and yield on large-diameter SiC remains a genuine engineering challenge.

---

Reliability, Qualification and Standards

Wide-bandgap devices fail differently from silicon, and the qualification frameworks have had to catch up.

JEDEC's JC-70 Wide Bandgap Power Electronic Conversion Semiconductors committee, with subcommittees JC-70.1 (GaN) and JC-70.2 (SiC), has published a substantial body of guidance. Key documents include:

• JEP180 — switching reliability evaluation for GaN power conversion devices
• JEP183 — threshold voltage measurement for SiC MOSFETs
• JEP184 — bias temperature instability evaluation for SiC MOS devices
• JEP186 — transient off-state withstand voltage robustness indicators for lateral GaN devices
• JEP187 — representing switching losses of SiC MOSFETs in datasheets
• JEP192, JEP194 and JEP195 (published 16 February 2023) — gate charge test method, gate oxide reliability evaluation, and gate switching instability evaluation for SiC devices

Alongside these sit ECPE AQG 324, the automotive power module qualification guideline widely used in European supply chains, and IEC 63373 for dynamic on-resistance characterisation of GaN devices.

Three failure mechanisms deserve particular design attention:

Threshold voltage instability in SiC MOSFETs. The SiC/SiO₂ interface contains more traps than its silicon equivalent, causing threshold voltage to shift under gate bias and temperature stress. This is why measurement methodology needed standardising: inconsistent measurement produced inconsistent datasheets and unfair comparisons between vendors.

Dynamic on-resistance in GaN HEMTs. After a high-voltage off-state period, charge trapping can temporarily raise on-resistance above the static datasheet value. If a designer sizes thermal budget from static R_DS(on) alone, real losses may exceed predictions.

Short-circuit withstand time. SiC MOSFETs generally tolerate short circuits for shorter durations than silicon IGBTs — often a few microseconds versus roughly ten. Protection circuits designed for IGBT timing may not act fast enough.

---

What Changes at the Board Level

The most frequent disappointment in wide-bandgap projects is a design that adopts the device but not the design discipline it requires. Fast switching is only an advantage if the surrounding circuit can tolerate it.

Gate drive becomes precision engineering. SiC MOSFETs typically need a specific positive gate voltage for low on-resistance and often a negative off-state bias to prevent parasitic turn-on. GaN HEMTs frequently have narrow gate voltage windows with little margin between correct operation and gate damage.

Layout parasitics dominate. At switching edges measured in single-digit nanoseconds, a few nanohenries of loop inductance produce voltage overshoot that can exceed the device rating. Commutation loops must be minimised and, ideally, laid out to cancel field.

EMI shifts upward in frequency. Faster edges move emissions into ranges where conventional filters are less effective and where board coupling paths that were previously negligible become significant. Many teams find EMI compliance, not efficiency, is the schedule risk.

Packaging becomes a first-order variable. A superb die in a package with high thermal resistance and high parasitic inductance will underperform a modest die in an excellent package. This is why top-side cooling, double-sided cooling and low-inductance module designs have received such intense development effort.

Figure 2 — Design constraint map for wide-bandgap converters Purpose: Show that device selection is one node in a coupled system, not an isolated choice. Layout: Central node labelled "WBG device selection (SiC / GaN)." Six nodes arranged radially, each connected bidirectionally to the centre and to their nearest neighbours: Gate drive design; PCB / module layout parasitics; Thermal management; EMI and filtering; Protection and fault response; Cost and supply assurance. Annotations: Label each spoke with its key parameter — gate drive: "dV/dt, Miller turn-on"; layout: "loop inductance (nH)"; thermal: "R_th junction-to-case"; EMI: "conducted and radiated emissions"; protection: "short-circuit withstand time"; cost: "$/A, second-source availability." Visual hierarchy: Centre node largest; use one accent colour for the centre and a neutral palette for spokes. Draw neighbour-to-neighbour links thinner than spokes to indicate secondary coupling. Caption: "Switching faster changes every adjacent design constraint. Teams that treat wide-bandgap adoption as a drop-in replacement typically rediscover this during EMI testing."

---

Latest Developments

Items below are dated and characterised by evidence type.

• 30 June 2025 — Wolfspeed Chapter 11 filing (established fact). Filed in the U.S. Bankruptcy Court for the Southern District of Texas under a prepackaged restructuring supported by a large majority of noteholders.
• 1 August 2025 — Munich Regional Court ruling (established fact, first instance). The court issued a first-instance ruling in Infineon's favour against Innoscience regarding a GaN patent in Germany.
• 29 September 2025 — Wolfspeed emerges from Chapter 11 (established fact). Debt reduced approximately 70%; maturities extended to 2030; annual cash interest expense down roughly 60%.
• October 2025 — onsemi begins sampling vertical GaN-on-GaN devices (company announcement). 700 V and 1,200 V parts from its Syracuse, New York facility, targeting 800 V DC-DC conversion. Vertical GaN would, if it scales, remove GaN's voltage ceiling — but sampling is not production.
• December 2025 and June 2026 — Enphase GaN microinverters (product shipment). IQ9 commercial units in December 2025, IQ9N residential units in June 2026, moving GaN into volume solar.
• 16 March 2026 — Texas Instruments 800 VDC reference architecture (product announcement). A complete architecture built to NVIDIA's 800 VDC reference design, shown at GTC 2026.
• 16 March 2026 — Navitas 800 V–6 V power delivery board (product announcement, targets not independently verified). Eliminates the intermediate 48 V bus converter stage; targeted 96.5% peak efficiency at 1 MHz.
• 27 May and 12 June 2026 — Innoscience v. Infineon in China (established legal fact; commercial impact disputed). The Suzhou Intermediate People's Court found infringement of two Innoscience GaN patents and awarded RMB 10 million (approximately $1.4 million) in damages, with injunctions. China's Supreme People's Court rejected Infineon's applications for reconsideration on 12 June 2026. Infineon has publicly disagreed with the decision, stated that it affects only a small subset of its GaN portfolio, and indicated it will pursue further legal options.
• Reported for July 2027 — TSMC GaN foundry exit (reported industry plan). TSMC is reported to be exiting GaN foundry services, with Navitas reported to be transitioning to Powerchip and GlobalFoundries. This reshapes fabless GaN supply chains and is worth monitoring rather than treating as settled.

---

Ultra-Wide Bandgap: Promising Science, Pre-Commercial Engineering

Beta-phase gallium oxide (β-Ga₂O₃), aluminium nitride and diamond are collectively described as ultra-wide-bandgap (UWBG) materials. Their theoretical merit is not in dispute: β-Ga₂O₃ has a bandgap near 4.8 eV and a theoretical critical field around 8 MV/cm, giving a Baliga figure of merit several times that of SiC or GaN. Gallium oxide has an additional structural advantage — it can be grown from a melt using techniques related to those used for silicon, which offers a potentially cheaper path to large native substrates than SiC's sublimation process.

Three obstacles keep these materials in the research phase:

1. Thermal conductivity. β-Ga₂O₃ conducts heat roughly an order of magnitude worse than silicon and far worse than SiC. A device that blocks enormous voltage but cannot expel its own heat is of limited practical use. Research into heterogeneous integration — bonding Ga₂O₃ to SiC or diamond — addresses this but adds process complexity.
2. P-type doping. Effective p-type doping of Ga₂O₃ remains unsolved, limited by low acceptor solubility, hole self-trapping and self-compensation. Without it, the bipolar device structures that underpin much of power electronics are unavailable.
3. Manufacturing maturity. Substrate diameters, supplier count and qualification data are all far behind SiC and GaN.

Progress is genuine. A March 2026 paper in Nature Communications reported a packaged, multi-die UWBG power module at the megawatt level for pulsed-power applications — a meaningful step, since previous UWBG demonstrations had been discrete devices at kilowatt scale.

Realistic assessment: UWBG materials are best understood as a research field with a plausible path to niche commercial applications — pulsed power, extreme environments, specialised RF — over the coming decade. Claims that they will displace SiC or GaN in mainstream power conversion in the near term are not supported by current evidence.

---

Common Myths and Frequent Mistakes

Myth: GaN will replace SiC. They address different voltage and current regimes. Vertical GaN could eventually contest SiC's territory, but as of mid-2026 that is a sampling-stage technology, not a market reality.

Myth: Wide-bandgap devices are drop-in replacements. They are not. Gate drive, layout, protection and EMI strategy all require rework.

Myth: Higher efficiency always justifies the cost. In many industrial drive applications, silicon IGBTs remain the rational choice. The business case depends on duty cycle, energy price, thermal system savings and volume — not on peak efficiency alone.

Mistake: Sizing thermals from static on-resistance. Dynamic R_DS(on) in GaN and temperature-dependent on-resistance in SiC both raise real-world losses above naive calculations.

Mistake: Reusing IGBT protection timing. SiC's shorter short-circuit withstand time can render an inherited desaturation protection design inadequate.

Mistake: Single-sourcing on price alone. Recent bankruptcy, litigation and foundry exits have all disrupted supply. Second-source qualification is now a risk-management necessity, not a procurement nicety.

---

Practical Takeaways

For design engineers

1. Budget schedule time for EMI compliance explicitly — assume it will be harder than your last silicon design.
2. Simulate with parasitic extraction from the actual layout, not idealised device models.
3. Verify dynamic on-resistance under your real switching conditions rather than trusting static datasheet values.
4. Confirm short-circuit withstand time against your protection circuit's actual response latency, measured rather than assumed.

For engineering managers and technical leads 5. Treat gate driver and packaging selection as first-order decisions made alongside the device, not afterwards. 6. Require vendors to state test conditions behind every efficiency and power-density claim. 7. Qualify a second source for any wide-bandgap part on a critical path.

For procurement and strategy 8. Scope market data before citing it; establish whether figures cover devices, the value chain or all industrial SiC. 9. Assess supplier financial health and litigation exposure alongside technical merit — both have disrupted supply in the past eighteen months. 10. Where 800 VDC infrastructure is relevant, engage suppliers on 2027 readiness now; qualification cycles are long.

---

Key Insights

1. The wide-bandgap advantage derives from critical electric field, not bandgap directly — thinner, more heavily doped drift regions mean lower resistance at a given blocking voltage.
2. SiC's vertical structure suits high voltage; GaN's lateral structure suits high frequency. This divides the market more reliably than any performance ranking.
3. Technical adoption and supplier profitability have diverged sharply; commercial success is not guaranteed by technical leadership.
4. The 800 VDC data centre transition is the first socket where SiC and GaN compete directly at volume.
5. Published market forecasts diverge by more than a factor of two, largely for definitional reasons.
6. Qualification standards have matured considerably through JEDEC JC-70, AQG 324 and IEC 63373 — a strong signal of industrial maturity.
7. Threshold voltage instability, dynamic on-resistance and short short-circuit withstand times are the three reliability issues most likely to affect a design.
8. Packaging and layout now determine whether theoretical device advantage survives into system performance.
9. Ultra-wide-bandgap materials are scientifically credible and commercially distant, constrained by thermal conductivity and p-type doping.
10. Supply chain concentration, patent litigation and foundry exits have made second-sourcing a core engineering requirement.

---

Frequently Asked Questions

What is a wide-bandgap semiconductor? A semiconductor whose bandgap substantially exceeds silicon's 1.12 eV — principally silicon carbide (~3.26 eV) and gallium nitride (~3.4 eV). The wider bandgap allows the material to withstand much stronger electric fields, enabling thinner, lower-resistance power devices.

Is SiC better than GaN? Neither is universally better. SiC dominates above roughly 900 V, particularly in EV traction inverters and grid converters. GaN dominates at 650 V and below where switching frequency and power density matter most, such as chargers and in-rack DC/DC conversion.

Why are SiC devices still more expensive than silicon? SiC crystal growth requires sublimation above 2,000 °C, wafer processing is slower and harder, and the industry is still transitioning from 150 mm to 200 mm wafers. Analysts estimate a current premium of roughly three to five times silicon.

When will SiC reach cost parity with silicon IGBTs? Analyst projections commonly point to the late 2020s for narrowing to roughly 1.5–2× on a per-device basis, driven by 200 mm wafer adoption. These are forecasts contingent on yield improvements, not commitments.

What is dynamic on-resistance? A temporary increase in a GaN HEMT's on-resistance after high-voltage off-state operation, caused by charge trapping. It can make real losses exceed those calculated from static datasheet values, which is why IEC 63373 exists.

Why did Wolfspeed file for bankruptcy if SiC demand is growing? The company had built substantial capacity funded by debt, then faced production delays, intensified Chinese competition and slower-than-forecast EV demand. It emerged from Chapter 11 on 29 September 2025 with roughly 70% less debt and its manufacturing footprint intact.

What is the 800 VDC data centre transition? A shift in AI data centres from 54 VDC in-rack power distribution to 800 VDC, reducing conversion stages and copper mass at megawatt rack scale. It relies on SiC and GaN devices, with volume deployment generally anticipated around 2027.

Can GaN be used in EV traction inverters? Not yet in production vehicles. Commercial lateral GaN's voltage ceiling makes it unsuitable for 800 V traction. Vertical GaN devices sampling since late 2025 could change this, but no automaker has announced a production GaN traction inverter.

What standards govern wide-bandgap device qualification? JEDEC's JC-70 committee publishes SiC and GaN test guidelines including JEP180, JEP183, JEP184, JEP186 and JEP187. ECPE AQG 324 covers automotive power module qualification, and IEC 63373 addresses GaN dynamic on-resistance.

What is gallium oxide and will it replace SiC? β-Ga₂O₃ is an ultra-wide-bandgap material with a bandgap near 4.8 eV and a very high theoretical figure of merit. Poor thermal conductivity and unsolved p-type doping keep it pre-commercial. Near-term displacement of SiC is not supported by current evidence.

Do wide-bandgap devices require different gate drivers? Yes. SiC MOSFETs typically require specific positive and often negative gate voltages, and GaN HEMTs have narrow gate voltage windows. Silicon gate drivers are generally unsuitable.

Why is EMI harder with wide-bandgap designs? Faster switching edges push emissions to higher frequencies where conventional filtering is less effective and parasitic coupling paths become significant. Many teams find EMI compliance is the dominant schedule risk.

What is the difference between 150 mm and 200 mm SiC wafers? A 200 mm wafer yields roughly 1.7–1.8 times more die than a 150 mm wafer at comparable defect density, spreading fixed processing costs across more devices. Analysts estimate 30–40% per-die cost reduction once yields mature.

Is SiC used outside electric vehicles? Yes — solar and wind inverters, industrial power supplies, rail traction, aerospace, energy storage, EV charging infrastructure and data centre power conversion. Note also that most "silicon carbide" by tonnage is abrasive and refractory material unrelated to semiconductors.

How reliable are wide-bandgap devices compared with silicon? Automotive-qualified SiC devices have accumulated substantial field experience and meet demanding qualification standards. Silicon still has decades more field data. The relevant question for any design is whether qualification data matches your specific mission profile.

---

Glossary

---

References

Academic papers and peer-reviewed research

Higashiwaki, M., & Wong, M. H. (2024). Beta-gallium oxide material and device technologies. Annual Review of Materials Research, 54(1), 175–198. https://doi.org/10.1146/annurev-matsci-080921-104058

Green, A. J., Speck, J., Xing, G., Moens, P., Allerstam, F., Gumaelius, K., … Higashiwaki, M. (2022). β-Gallium oxide power electronics. APL Materials, 10(2), 029201. https://doi.org/10.1063/5.0060327

Sun, J., & Zhang, J. (2025). Recent advanced ultra-wide bandgap β-Ga₂O₃ material and device technologies. Advanced Electronic Materials, 11, 2300844. https://doi.org/10.1002/aelm.202300844

Pavlidis, S., Medwig, G., & Thomas, M. (2024). Ultrawide-bandgap semiconductors for high-frequency devices. IEEE Microwave Magazine, 25(10), 68–79. https://doi.org/10.1109/MMM.2024.3428193

A megawatt ultra-wide bandgap semiconductor module for pulsed power electronics. (2026). Nature Communications. https://www.nature.com/articles/s41467-026-71274-6

Wide and ultrawide bandgap power semiconductors: A comprehensive system-level review. (2026). Electronics, 15(4), 835. https://doi.org/10.3390/electronics15040835

A comprehensive study on GaN power devices: Reliability, performance, and application perspectives. (2025). Electronics, 14(22), 4430. https://doi.org/10.3390/electronics14224430

Standards and technical guidelines

JEDEC Solid State Technology Association. (2021). JEP183: Guidelines for measuring the threshold voltage (V_T) of SiC MOSFETs. https://www.jedec.org/standards-documents/docs/jep183

JEDEC Solid State Technology Association. (2021). JEP184: Guideline for evaluating bias temperature instability of silicon carbide metal-oxide-semiconductor devices for power electronic conversion. https://www.jedec.org/standards-documents/docs/jep184

JEDEC Solid State Technology Association. (2021). JEP180.01: Guideline for switching reliability evaluation procedures for gallium nitride power conversion devices. https://www.jedec.org/standards-documents/docs/JEP180

JEDEC Solid State Technology Association. (2023, February 16). JEDEC wide bandgap power semiconductor committee publishes a series of documents for reliability and testing of silicon carbide (SiC) MOSFETs [Press release]. https://www.jedec.org/news/pressreleases/jedec-wide-bandgap-power-semiconductor-committee-publishes-series-documents

JEDEC Solid State Technology Association. (n.d.). Wide bandgap power semiconductors (GaN, SiC) technology focus area. https://www.jedec.org/category/technology-focus-area/wide-bandgap-power-semiconductors-gan-sic

Company documentation and announcements

NVIDIA. (2026). NVIDIA 800 VDC architecture will power the next generation of AI factories. NVIDIA Technical Blog. https://developer.nvidia.com/blog/nvidia-800-v-hvdc-architecture-will-power-the-next-generation-of-ai-factories/

Texas Instruments. (2026, March 16). TI unveils complete 800 VDC power architecture for future generation AI data centers with NVIDIA [Press release]. https://www.ti.com/about-ti/newsroom/news-releases/2026/2026-03-16-ti-unveils-complete-800-vdc-power-architecture-for-future-generation-ai-data-centers-with-nvidia.html

STMicroelectronics. (2026, June 29). Update: 800 V HVDC for AI data centers thanks to 6 kW, 12 kW, and 20 kW power delivery boards. The ST Blog. https://blog.st.com/800-v-hvdc-data-center/

Navitas Semiconductor. (2025, October 13). Navitas supports 800 VDC power architecture for NVIDIA's next-generation AI factory computing platforms [Press release]. https://navitassemi.com/navitas-supports-800-vdc-power-architecture-for-nvidias-next-generation-ai-factory-computing-platforms/

Wolfspeed, Inc. (2025, September 29). Wolfspeed successfully completes financial restructuring, emerges as financially stronger company well positioned in silicon carbide market [Press release]. https://www.wolfspeed.com/company/news-events/news/wolfspeed-successfully-completes-financial-restructuring-emerges-as-financially-stronger-company-well-positioned-in-silicon-carbide-market/

Wolfspeed, Inc. (2025, September 8). Wolfspeed's plan of reorganization confirmed, clearing path to emerge from restructuring process as a financially stronger company [Press release]. https://investor.wolfspeed.com/news/news-details/2025/Wolfspeeds-Plan-of-Reorganization-Confirmed-Clearing-Path-to-Emerge-from-Restructuring-Process-as-a-Financially-Stronger-Company/default.aspx

Infineon Technologies AG. (2025, July 3). Infineon advances on 300-millimeter GaN manufacturing roadmap as leading integrated device manufacturer [Press release]. https://www.infineon.com/press-release/2025/INFXX202507-122

Innoscience. (2026, June 12). Infineon found to infringe Innoscience patents, barred from selling in China [Press release]. https://www.innoscience.com/news/press-releases/11594

Industry reports and analysis

Yole Group. (n.d.). Power SiC — Markets and applications. https://www.yolegroup.com/product/report/power-sic-2025---markets-and-applications/

IDTechEx. (2026). Power electronics market 2026–2036: Data centers, electric vehicles, and renewables. https://www.idtechex.com/en/research-report/power-electronics-market/1152

IDTechEx. (2025). Power electronics for electric vehicles 2026–2036: Technologies, markets, and forecasts. https://www.idtechex.com/en/research-report/power-electronics-for-electric-vehicles/1125

Mesago / PCIM. (2026). SiC vs GaN in 2026 and beyond. PCIM Expo News. https://news.pcim.mesago.com/

PatSnap. (2026). SiC power device technology landscape 2026. https://www.patsnap.com/resources/blog/articles/sic-power-device-technology-landscape-2026/

News and legal reporting

Semiconductor Today. (2026, June 15). China's Supreme Court upholds injunction against Infineon, says Innoscience. https://www.semiconductor-today.com/news_items/2026/jun/innoscience-infineon-150626.shtml

South China Morning Post. (2026, June 15). Chinese compound chip stocks surge after Supreme Court blocks Infineon in GaN patent case. https://www.scmp.com/tech/tech-trends/article/3357172/

Data Center Dynamics. (2026). Nvidia prepares data center industry for 1MW racks and 800-volt DC power architectures. https://www.datacenterdynamics.com/en/news/nvidia-prepares-data-center-industry-for-1mw-racks-and-800-volt-dc-power-architectures/

Sourcing note. Material-property values in Table 1 are consensus figures from standard device-physics literature rather than any single proprietary source. Market forecasts are attributed to their originating firms and should be read as estimates with disclosed methodological limits. Vendor performance claims are identified as such throughout and have not been independently verified by OneWise.

---