Two companies are starting sales of entire chipmaking facilities that will cost as little as $30 million and occupy a space that is tiny compared to a typical multibillion-dollar chip fab.
The mini fabs will make lagging-edge power semiconductors and flexible chips for retail labeling and consumer electronics. The facilities can be built in locations once thought impractical for chip production, the company CEOs told EE Times in interviews.
U.K.-based Pragmatic says its “Fabs-in-a-Box” will cost orders of magnitude less than the typical chip fab with a $20 billion price tag, and they can start production in about a third of the time.
ADVERTISING
“We can look at building a new line out in as little as 12 to 14 months versus the typical three to five years,” noted Pragmatic CEO David Moore. “Because we’re using fewer steps and more cost-effective materials, we’re able to get to a more optimized cost overall for the product.”Across the Atlantic in the U.S., Nanotronics has demonstrated that its “Cubefabs” can make devices that outperform some state-of-the-art power semiconductors.
“We’re achieving higher kilovolts while maintaining the thermal properties,” Nanotronics CEO Matthew Putman told EE Times. “We’re now making these chips and collaborating with semiconductor companies to scale their production.”
Prefabricated fab
Nanotronics’ Cubefab is a modular, prefabricated factory designed to work in 85% of the world’s environments—except some seismically tricky spots. While the mini fab has production tools like those typically used to make power semiconductors, the company says it has used AI tech to deliver superior results with gallium oxide, an ultra-wide bandgap material.
The key is Nanotronics’ use of generative AI, according to Putman.
“That’s why we can produce chips like gallium oxide—our first product, though we can do others,” Putman said. “It’s about taking something scientifically proven but not yet scalable and using AI to boost yields and make it cost-competitive, all while being several times better than what’s out there.”
There is no need in a Cubefab for ASML equipment or other costly tools typical chipmakers rely on, Putman said.
“We’ve built strong partnerships with equipment suppliers, and our own algorithms and gear give us confidence.”
The company is building a pilot line in New York that will open later this year.
Different backgrounds
The creators of mini fabs come from different backgrounds. Founded in 2010, Nanotronics sells inspection tools and AI-driven process controls. Started around the same time, Pragmatic is making its first flexible ICs for near-field communications and radio-frequency identification in retail labels and healthcare wearables.
Nanotronics CEO Matthew Putman. (Source: Nanotronics)
“We do that by bringing together thin-film transistor technology with state-of-the-art 300-millimeter wafer processing,” Moore said. “That’s never been done before.”
Pragmatic prints circuitry on a flexible thin film that Moore says is entering new applications where rigid silicon does not work.
“Some of these layers are five to 10 nanometers thick,” Moore says. “We’re bringing together thousands of TFTs per square millimeter. While it’s not at the five-nanometer advanced node level, it is actually some very complex integrated processing.”
The company started its first 300-mm facility last year and is ramping up another that will enter production by the middle of 2025.
Pragmatic’s 300-mm wafer. (Source: Pragmatic)
“By the end of the year, we will be at an annual capacity of billions of flex ICs per year,” Moore said. “There is a very strong interest among our customers in being able to localize manufacturing right with them on site. In those situations, the cost of the fab will depend on where it is, what exact tools are going into it, but the fundamentals of the tooling and the recipe are consistent.”
Cubefab
A Cubefab can cost as little as $30 million to start production, Putman says. A central cube connects up to four “petals” that are the production facilities.
“The central cube houses shared resources, and each petal is a fab,” Putman added. “Each petal costs between $30 and $40 million, depending on specs. You can start with one, and adding more gets cheaper. The beauty is production never stops. Since the petals are separate, you can build another without interrupting operations, and the Cubefab network lets you share capacity.”
Fab-in-a-Box
Pragmatic has built several Fabs-in-a-Box on its production site in Durham, U.K. One fab is 20 meters by 30 meters, yet capable of making billions of flex ICs per year, Moore said.
“It’s extremely capital efficient, but very space and production efficient as well. It’s extremely sustainable. We go from raw materials to wafers out in just days rather than months for standard silicon. We’re also using a lot less energy, a lot less water, and that means that these chips have an order-of-magnitude lower carbon footprint too.”
Pragmatic aims to offer existing customers its Fabs-in-a-Box to extend their product lines. The company is starting sales of chips under the Pragmatic brand while providing foundry services to customers capable of designing their own products.
“We’ll see probably about a 60-40, 70-30 split between IDM and foundry,” Moore said. Longer term, he expects the foundry share and sales of Fabs-in-a-Box to grow.
“Localization of domestic sources of semiconductors is a very strategic element today in the world. It’s rare that you can build a fab in such little time, so carbon efficiently, but still have the kind of volumes that we can pump out: 600 square meters, a billion flex ICs per year, it really is a very interesting and disruptive approach.”
Potential buyers
Putman will not name buyers for the Cubefabs yet.
“They’re semiconductor firms exploring power electronics divisions to complement their traditional silicon offerings, or they’re already in the compound semiconductor space, looking to leapfrog silicon carbide,” he says. “If we break ground on a Cubefab for someone this year, we’d be thrilled. But the real milestone is shipping these chips and seeing them work in data centers or power inverters for electric vehicles. We’re confident they’ll be a drop-in replacement for existing solutions.”
Putman also says it is possible to build Cubefabs in locations where chips have never been made.
“I’m very excited about places like Africa and the Global South,” he said. “There’s incredible AI-native talent there, but the semiconductor game has been too expensive for them to enter. These regions would never consider a multibillion-dollar fab, but this is different.”
Nanotronics makes inspection tools yet sees its main innovations in using factory data to improve yields and performance with advanced computation.
“Our AI models have been learning for 15 years, even before we had the AI to process it,” Putman said. “That data is key to running efficient fabs. It’s all connected—using data to make things better. With Cubefabs, we’re now making the final product ourselves.”
