Gallium Nitride and Gallium Oxide

Silicon carbide and gallium nitride crossed from laboratory curiosity to qualified production part in under two decades. SiC now anchors 800V electric vehicle traction inverters and DC fast chargers. GaN-on-silicon anchors USB Power Delivery chargers and AI server power supplies. The interesting question in 2026 is no longer whether wide bandgap works. It is where the ladder goes next, and how much of the next rung is real.

In this piece I take the forward view. I look at the industry-wide economics of the 200mm SiC transition, at how GaN reaches beyond power conversion into RF and defence, at vertical GaN, and then at gallium oxide, the ultra-wide-bandgap material most often named as the successor at extreme voltages. Gallium oxide is where I have to be most careful. The research I draw on documents it as a research direction, not an industry, and I keep that boundary visible. I close with the supply-side layer beneath all of these materials, because that is where the shared risk sits.

Why the bandgap ladder keeps mattering

Bandgap energy is the single property that shapes most of what a power semiconductor can do. A wider bandgap raises the critical electric field at which the material breaks down. A higher breakdown field allows a thinner drift layer for a given voltage rating. A thinner drift layer lowers on-state resistance and conduction loss for the same device footprint. That chain of consequences is the whole case for wide bandgap.

The numbers anchor it. Silicon has a bandgap of 1.1 eV. Silicon carbide, in its 4H polytype, is 3.26 eV. Gallium nitride is 3.4 eV. Both wide bandgap materials sit roughly three times wider than silicon. The breakdown field figures follow the same direction: silicon at 0.3 MV/cm against SiC at 2.8 MV/cm. A tenfold difference in critical field is why an 800V inverter can shed a meaningful share of its conduction losses when it moves from silicon to SiC.

The relationship between bandgap and breakdown field is not linear. It compounds, so each further step up the ladder buys a disproportionately larger gain in the voltage a device of a given thickness can withstand. That compounding is the economic logic behind ultra-wide-bandgap research. If a material offers a critical field meaningfully higher than SiC's, it promises a comparable step change at voltage classes where SiC itself begins to strain: multi-kilovolt grid-tie inverters, high-voltage DC transmission, and pulsed-power applications where a device would otherwise have to be stacked in series to survive.

The term I need is ultra-wide bandgap, or UWBG. Materials science draws the boundary loosely at a bandgap noticeably wider than GaN's. The category in broader academic literature includes gallium oxide, synthetic diamond, and aluminium nitride, each at a very different stage of maturity. That taxonomy is safe to state. What is not safe is attaching market sizes, company names, or production-readiness claims to any UWBG material without a primary source, and I do not.

The 200mm SiC transition is an industry migration

Company-level coverage of SiC often treats the move from 150mm to 200mm wafers as a single firm's bet. Wolfspeed's case is the best known. It involved a combined investment of roughly $6 billion across its Mohawk Valley and Siler City facilities, and a company-specific target of a 50% per-die cost reduction. But the transition is not a Wolfspeed event. According to the compound-semiconductor fab reference I draw on, SiC fabrication today runs across both 150mm and 200mm planar and trench process lines at Infineon, STMicroelectronics, Wolfspeed, ROHM, and onsemi.

Each of those companies is making the same underlying economic bet. A 200mm wafer has a larger usable area. If the cost of growing a SiC crystal scales more slowly than the area it produces, the per-die cost falls, and that fall has to be large enough to justify requalifying an entire fab line around a new wafer diameter. The technical reference puts the industry-wide target at a 20% to 30% reduction in per-die cost.

The gap between that figure and Wolfspeed's own 50% target is informative. My reading is that Wolfspeed's fully automated, purpose-built 200mm line is aiming to beat what the wider industry is likely to realise, because most of the industry is layering 200mm onto lines that were not designed around it from the ground up. That is an interpretation, not a sourced statement, and I flag it as one. The two numbers themselves are sourced.

What this means for a buyer

If you buy SiC devices or plan around them, the useful framing is that cost convergence with silicon at a given voltage class will be set by the aggregate pace of many requalification exercises, not by any one fab announcement. The merchant substrate suppliers, Wolfspeed, Coherent, ROHM's SiCrystal, STMicroelectronics' Norstel, and SK Siltron CSS, and the device makers drawing on them are all on their own timelines. A delay at one shifts the average but does not stop the migration.

GaN beyond power conversion

Most coverage of GaN is about chargers, and that is the power-electronics face of the material. Its device architecture, the high electron mobility transistor built on the two-dimensional electron gas that forms at the AlGaN/GaN heterojunction, was first commercially proven in RF power amplification, not in power conversion. That lineage never went away. It simply runs on a different substrate.

The substrate choice follows the design driver. Power GaN rides silicon, because that lets it use existing silicon fab infrastructure and keeps cost per watt low. RF GaN increasingly rides silicon carbide. GaN-on-SiC gives up the lower substrate cost of silicon in exchange for better lattice matching and higher thermal conductivity. When the application is a 5G base-station power amplifier or a defence radar transmit-receive module dissipating heat from a dense antenna array, heat matters more than cost per watt.

The fab reference I use makes the split explicit. It lists GaN applications as 40V to 650V power switches for USB PD fast chargers and AI server power supplies, alongside 5G base-station RF power amplifiers. One material system, two application families, served by two substrate choices.

Not displacing GaAs or InP

It is tempting to read GaN's RF growth as GaN displacing everything else. The source does not support that. Gallium arsenide remains the incumbent for handset RF power amplifiers, supplied at scale by companies including Win Semiconductor, Qorvo, and Skyworks, grown by the LEC process on predominantly 150mm wafers. Indium phosphide sits further out in frequency and serves telecom lasers and RF above 100 GHz, a regime beyond where GaN-on-SiC currently competes.

The better description is that GaN is carving out the mid-to-high-power, mid-frequency territory: 5G infrastructure, radar, and electronic warfare. There its combination of high breakdown voltage and high electron mobility beats GaAs on power handling, without needing InP's more specialised and lower-volume fabrication base.

Vertical GaN: the furthest-out piece

Nearly every GaN device shipping today is lateral. Current flows horizontally across the heterojunction, parallel to the wafer surface. That geometry is what made GaN-on-silicon commercially viable, because it maps onto planar CMOS-style processing.

Vertical devices route current through the thickness of the material, the same basic geometry silicon and SiC power devices already use. In principle it scales more favourably to higher voltage and current density, because blocking capability comes from bulk thickness rather than lateral surface distance. It is a real, active research area aimed at pushing GaN above the voltage classes where lateral GaN-on-silicon competes, potentially into territory SiC holds today.

My sources describe the lateral architecture in detail but do not document vertical GaN's commercialisation timeline, named developers, or production status. I mark all three as unsourced. If a vendor tells you vertical GaN is imminent, the useful reply is to ask for the qualification data.

Gallium oxide: a promising material and a genuine evidence gap

This is the section where I write differently, and I would rather say so than let the change of register pass. Everywhere else I can point to a named technical reference or fab specification. For gallium oxide I cannot. The compound-semiconductor and wide-bandgap references I consulted cover SiC, GaN, GaAs, and InP in depth and do not mention Ga2O3 at all. I treat that silence as a finding, not an oversight to route around.

What is well established

Gallium oxide, most often studied in its beta phase, is an ultra-wide-bandgap semiconductor with a bandgap larger than both silicon carbide's and gallium nitride's. Following the same physics that lets SiC and GaN outperform silicon, a wider bandgap implies a higher critical breakdown field, thinner drift layers, and lower conduction losses at voltages where even SiC needs thicker, more resistive structures.

It is consistently discussed as most promising for extreme-voltage power electronics, well above the 800V to 3.3kV range where SiC already competes. It is not generally framed as a near-term replacement for GaN in the lower-voltage, high-frequency applications GaN dominates. The better picture is an eventual complement further up the voltage ladder.

What I will not state

Beyond that, everything is unsourced in my base. That includes Ga2O3's precise bandgap and breakdown field, since public academic literature reports specific numbers but I have no primary reference to stand behind one. It includes the current or projected market size for Ga2O3 power devices, named companies making substrates or devices at any scale, whether wafers exist at research, pilot, or commercial volume, and the voltage classes at which any device has been qualified as opposed to demonstrated in a lab. It also includes any India-specific activity.

The one inference that follows from chemistry

There is one safe inference that does not need a Ga2O3 market report. Gallium oxide is a compound of gallium. Whatever upstream concentration affects gallium as a raw material affects Ga2O3 exactly as it affects gallium nitride. Both draw on the same elemental feedstock, refined through the same chain, however differently they are later processed into wafers.

My book's chapter on semiconductor materials documents that chain. More than 98% of primary low-purity gallium is produced in China as a byproduct of bauxite refining, and China imposed formal export controls on gallium in August 2023. A future in which Ga2O3 moves from research material to qualified device inherits that chokepoint on the first day. It is not a new risk to be discovered later. It is an already-documented one.

Reading that table left to right, the first three columns rest on the same technical references. The fourth rests on thinner ground, and the table says so. That asymmetry is the organising fact of this whole discussion. SiC and GaN are documented industries, with named suppliers, quoted parameters, and forecast market sizes. Ga2O3 is a documented research direction.

How I would use this when planning around these materials

For an engineer or sourcing lead, the practical value of the ladder is in matching each material to its real voltage and frequency territory rather than to its hype. Below about 600V, silicon still competes and GaN-on-silicon is taking the cost-driven, high-frequency conversion slots such as chargers and server power supplies. From 650V up to 3,300V and beyond, SiC is the qualified choice, and it is where the 200mm cost trajectory matters most. For RF at mid-to-high power, GaN-on-SiC is the material to evaluate, with GaAs still the default for handsets. Anything above that, or any claim about a material outside those ranges, needs a qualification record before it belongs in a design.

On the supply side, I would ask two separate questions for every wide-bandgap part. Who makes the substrate, and how many qualified sources exist? And where does the elemental feedstock come from? For SiC the first question dominates. For GaN the second one matters as well, because the gallium chain runs through a single country's refining capacity and a licensing regime that has tightened since 2023. A design that is dual-sourced at the device level can still be single-sourced at the mineral level.

The strategic layer: concentration below the device

Step back from any single material and the same pattern recurs. Manufacturing capability concentrates in a short list of companies and countries, and the concentration sits upstream of the device fab rather than in it.

For SiC, the chokepoint is the substrate. Wolfspeed alone is estimated to control 50% to 60% of global merchant SiC substrate supply, and it sells into its own device-level competitors. The remainder is split among Coherent, ROHM's SiCrystal, STMicroelectronics' Norstel, and SK Siltron CSS. The binding constraint is capital-intensive, defect-sensitive crystal growth built up over years. It is not a raw-material shortage.

For GaN, the chokepoint is more directly mineral. Gallium is concentrated above 98% in Chinese primary production, with germanium at 60% to 70% and both under Chinese export licensing since August 2023. That gallium feeds a GaN substrate and epitaxy industry that is itself concentrated among a handful of merchant and captive producers. Two layers of concentration sit on top of each other.

India's position

I will not invent a gallium oxide narrative for India where the sources have none. What the sources do show is two unresolved gaps that any future Indian position in wide or ultra-wide bandgap would sit on. First, the raw-material chain: India has near-total import dependence on electronic-grade gallium, germanium, indium, and high-purity polysilicon, and the 2023 critical-minerals response, the National Critical Minerals Mission, the 30-mineral list, and the Minerals Security Partnership, targets the mineral layer without yet closing the wafer or specialty-chemical gap above it. Second, the device layer: CDIL's SPECS-backed expansion into SiC discrete packaging in Mohali is a real but early-stage footprint, and domestic SiC substrate capacity is unsourced.

What to take away

The line I keep coming back to is that wide bandgap changed what a power device could survive, while ultra-wide bandgap is for now a bet that the same physics keeps paying dividends beyond the point where anyone has built a qualified commercial fab to prove it. Every rung of the ladder, proven or not, traces back to the same short list of substrate producers and mineral refiners. If I were planning around any of these materials, I would spend at least as much time on that upstream list as on the device datasheet.