In a lab in Saitama, north of Tokyo, a machine pulls a fat, dark crystal out of a pool of molten oxide the way a candlemaker pulls a taper out of wax. The crystal is gallium oxide. What makes it interesting is not the crystal. It is the pool of melt. Silicon is grown this way, cheaply, by the ton. Silicon carbide, the material that now runs the power electronics in every serious electric car, is not: you have to coax it out of vapor at 2,300°C, one slow wafer at a time, which is a large part of why a silicon carbide diode costs what it costs.

Gallium oxide grows like silicon. That single fact is why a small cluster of companies has spent a decade betting that it becomes the third material in power electronics, after silicon and silicon carbide. And it is why, if you read the patents instead of the press releases, the most important story in next-decade power chips is a three-way standoff where the United States holds the weakest hand.

The niche nobody outside the field is watching

Search the US grant record for gallium oxide and you get 1,369 issued patents, 151 of them since the start of 2024. That number is misleading in a way worth pausing on, because it is the exact trap these stories usually fall into: the large majority of those grants belong to a single Japanese company, Semiconductor Energy Laboratory, and they are not power chips at all. They are the indium-gallium-zinc-oxide transistors that switch the pixels in flat-panel displays. Same two words in the abstract, completely different invention.

Strip those out and read the claims of what remains, and a much smaller, much sharper picture appears: a few dozen patents that all describe the same physical object. A vertical stack of single-crystal gallium oxide, doped n-type, with a trench cut into it and a metal contact laid across the trench to spread out the electric field before it tears the device apart. Cornell’s “vertical gallium oxide power FETs.” TDK’s “Schottky barrier diode,” filed jointly with a company called Novel Crystal Technology. FLOSFIA’s “normally-off” switch with a threshold above 3 volts. Delete the phrase “gallium oxide” from every one of these and they are still unmistakably the same device, solving the same problem: how to make a switch that blocks a thousand volts in a sliver of material a fraction the thickness silicon carbide would need.

That is the prize. Gallium oxide’s bandgap is about 4.8 electron-volts in its stable form, and 5.3 in the version FLOSFIA grows. Silicon carbide is 3.3; gallium nitride, 3.4. A wider bandgap means the material tolerates a far stronger electric field before it breaks down, which means a device can be thinner, switch faster, and waste less energy as heat while holding back the same voltage. In an electric-vehicle inverter, the box that turns battery DC into the AC that spins the motor, less wasted energy is directly more range. In a data center, it is a smaller, cooler power supply feeding the racks of AI accelerators that are currently eating the grid.

The paper that started it

Almost all of it traces to one experiment. In 2012, Masataka Higashiwaki, a researcher at Japan’s National Institute of Information and Communications Technology, built the first working transistor out of a single crystal of gallium oxide, using substrates grown by Tamura Corporation. The paper, in Applied Physics Letters, has since been cited more than 1,500 times. His follow-on review, “Development of gallium oxide power devices,” is the document nearly everyone in the field builds on.

Higashiwaki did not stop at the paper. In 2015 his institute and Tamura spun out Novel Crystal Technology, which has since demonstrated 150-millimeter gallium oxide wafers grown from melt, the same diameter as a mainstream silicon carbide wafer, at a fraction of the crystal-growth cost. Novel Crystal makes the wafers; TDK, on the patents, makes the diodes on top of them. The two names appear on the same filings.

The other Japanese line runs through Kyoto. FLOSFIA, a university spin-out, chose a harder, metastable form of the crystal, grown by spraying a fine mist of precursor onto a hot surface, and used it to build what it says was the world’s first normally-off gallium oxide transistor. In 2018, Toyota’s main parts supplier, DENSO, put money into FLOSFIA’s roughly $7 million Series C and signed on to co-develop the devices for electrified vehicles. FLOSFIA now ships engineering samples of its diodes under the brand GaO, and announced a mass-production plan in December 2024. It named a new president, Takashi Shinohe, in March 2025.

The two things wrong with the miracle material

Read far enough into the claims and the reason this took a decade becomes obvious. Gallium oxide has two flaws that would each be disqualifying on their own.

First, it barely conducts heat. Its thermal conductivity is somewhere around 10 to 27 watts per meter-kelvin. Silicon carbide’s is roughly 490. A gallium oxide switch handling real power develops hotspots that the material cannot carry away, so half the engineering is about bonding the thin active layer onto a foreign substrate that can.

Second, and stranger, nobody has ever made it conduct the other way. Every semiconductor device you have ever used depends on pairing n-type material, which carries current with electrons, against p-type, which carries it with the absence of electrons. Gallium oxide will not go p-type. It resists every dopant tried. So the entire device zoo in these patents is a set of workarounds for a missing half of physics: trenches and field plates to manage the electric field without a p-n junction, and heterojunctions that borrow p-type behavior from a different oxide bolted on top. In October 2024, Toyota and DENSO were granted a patent on a “p-type gallium oxide semiconductor device with alternating layers,” an attempt to fake the missing polarity with a finely engineered stack. That a carmaker is filing on the deepest open problem in the material tells you how serious the automotive bet has become.

America’s hand: a university and the Navy

Here is where the map gets uncomfortable for a US reader. The commercial patents are Japanese. The leading American position is a single university group: Huili Grace Xing and Debdeep Jena at Cornell, working with materials scientist Darrell Schlom. Their lab built vertical gallium oxide transistors that block more than 1.6 kilovolts, and holds patents on the trench structures and on a novel crystal-growth method, “suboxide molecular-beam epitaxy,” that grows the films several times faster than the standard technique. In July 2025 the group reported a fix for a long-standing problem making good electrical contact to the material.

The work is funded largely by the US military. The Cornell effort runs through a center co-sponsored by the Air Force Research Laboratory, and the Naval Research Laboratory holds its own gallium oxide device patents. When the buyers of last resort for your national semiconductor bet are the Air Force and the Navy, the commercial industry has not arrived yet.

The element China owns

And then the part that turns an engineering story into a strategic one. Gallium the element is a byproduct of aluminum and zinc refining, and China produces something like 95 percent of the world’s supply. In August 2023, Beijing put gallium under export licensing. In December 2024 it banned gallium exports to the United States outright, then later suspended the ban while keeping the license requirement in place. The US Geological Survey estimated a full ban could cost the American economy on the order of $3.4 billion and push gallium prices up more than 150 percent.

So line up the three players. The cheap, melt-grown material that could undercut silicon carbide on cost is real. The patents that turn it into a shippable device are mostly held in Japan. The raw element it is made of is controlled almost entirely by China. The American position is a Cornell lab and two military research labs holding the door open. A power-electronics company deciding today where to place its next SiC-versus-something bet is not really choosing a material. It is choosing which of those three dependencies it can live with.

Method note. Patent counts come from the full text of 9.3 million US utility grants sourced from USPTO bulk grant data, searched for “gallium oxide” and filtered to issued grants; the 1,369 figure is the raw keyword count and, as noted in the piece, is dominated by display-transistor filings unrelated to power devices. The power-device subset was isolated by requiring claims to also describe Schottky diodes, drift layers, breakdown voltage, or power-semiconductor structures, and by reading the abstracts and claims of the leading filings directly. Assignee counts combine variant spellings and jointly assigned filings; the TDK, Tamura, and Novel Crystal figures overlap because those companies co-assign the same patents. Citation counts for the founding papers are from OpenAlex. Material properties, commercialization milestones, funding figures, and the gallium export-control timeline are drawn from NICT, Cornell, DENSO, and FLOSFIA announcements and from reporting by IEEE Spectrum, the Stimson Center, Fastmarkets, and the US Geological Survey. A skeptical reader should note that no gallium oxide power device is yet in mass production; the claims here are about patents filed and samples shipped, not products at scale.