Capital · United States

The U.S. Just Bet $1 Billion on Quantum Chip Manufacturing

Credited to IEEE Spectrum · spectrum.ieee.org · 11h ago

Photograph · IEEE Spectrum

The U.S. Department of Commerce has finalized a CHIPS and Science Act R&D award of up to $1 billion for Anderon, a quantum foundry IBM established earlier this year. The 16 September announcement backs a manufacturing service for IBM and other quantum-hardware developers.

IBM is matching the award with $1 billion in cash and contributing intellectual property, assets, and skilled workers. Headquartered in Albany, N.Y., Anderon builds on an existing operation at Albany NanoTech, where IBM has made quantum wafers for more than five years, says CEO Mukesh Khare. That existing infrastructure has helped Anderon get started: Its first wafers are already moving through a production-scale, 300-millimeter line.

Quantum computing “is moving from science to real commercial impact now,” Khare says. “That means the industry also needs a more stable manufacturing source for building quantum chips.”

If Anderon is successful, it could do for quantum hardware what TSMC did for conventional chips: let companies focus on designing chips without each one having to build a multibillion-dollar factory for the underlying hardware.

What an Automated Quantum Fab Can Offer

IBM says shifting much of its quantum wafer R&D from the 200-millimeter fabrication lab at the IBM Research headquarters in Yorktown Heights, N.Y., to the Albany 300-millimeter line, beginning in 2021, doubled chip-development speed and increased its chips’ physical complexity tenfold. Round-the-clock processing shortened fabrication cycles, and advanced manufacturing techniques enabled more elaborate wiring and qubit connections. Anderon draws on Albany NanoTech’s manufacturing capability and IBM’s experience deploying more than 90 quantum systems since 2016. The challenge now is extending those capabilities to outside designers. Anderon’s potential customers include processor developers without fabs, system builders, and national laboratories. Even companies with their own manual chipmaking facilities may be interested in outsourcing fabrication to an automated foundry. Rigetti Computing, for example, another CHIPS Act R&D award winner, is in talks with Anderon, according to CEO Subodh Kulkarni.

Rigetti currently makes processors on 150-millimeter wafers in its manual fab in Fremont, Calif. Kulkarni puts its yield near 50 percent, which he says is “good enough” for R&D when specs are changing frequently. But Anderon promises more reliable automated chip production that companies like Rigetti could retain to advance their own chip development. “We find that to be very attractive,” Kulkarni said, “not because of capacity, but because of capability.”

Like conventional silicon foundries, Anderon plans to offer process design kits, or PDKs, specifying the rules designers must follow to use its manufacturing processes. “They design, we build,” Khare says. Customization will also be available, he said, and the foundry will protect clients’ intellectual property and operate neutrally, including for IBM’s competitors.

Making that model work will require predictable fabrication. For superconducting qubits, a critical component is the Josephson junction, which consists of two superconductors separated by a thin insulating barrier. Variations in junction area and barrier thickness can change critical current and qubit frequency, potentially creating unwanted frequency overlaps and reducing chip yield.

Anderon’s automated line could help control variation through repeatable patterning, deposition, and oxidation, with in-line testing to identify process drift. But customers will need to measure the actual improvement in qubit performance, Kulkarni says.

Anderon’s latest-generation 300mm quantum wafers. Anderon

Much of the fabrication equipment overlaps with that used for silicon chips, although superconducting circuits require specialized materials and process steps. Anderon advertises superconducting qubit arrays, microwave components, and through-silicon connections. But extending manufacturing capabilities to other quantum technologies, like trapped ion or photonic qubits, would require additional process development that Anderon has not publicly detailed yet.

Semiconductor spin qubits can already be fabricated on conventional chip production lines, says Lieven Vandersypen of Delft University of Technology in the Netherlands, while superconducting qubits require specialized steps that may require new development.

A dedicated quantum facility could spread costs and support the development of processes optimized for quantum devices, according to Vandersypen. “Understanding how to make this transition is timely and important,” he says.

Manufacturing capabilities alone will not make Anderon quantum computing’s TSMC. It also needs sustained demand beyond IBM. Designers have alternatives. Vandersypen points to QuantWare’s investment in its own wafer-scale facility and Belgium’s imec, which supplies prototype quantum chips rather than pursuing high-volume manufacturing. “Time will tell which model will prevail. Perhaps both will co-exist,” he says.

Commerce is backing several approaches to strengthen U.S. competitiveness in the global quantum market. While Anderon’s award is the largest so far, Commerce has finalized up to $375 million for GlobalFoundries and up to $100 million each for Rigetti, D-Wave, Quantinuum, and PsiQuantum. Those awards support manufacturing for multiple quantum architectures and address bottlenecks including readout electronics, cryostats, superconducting materials, optical components, and photonic packaging. Shared wafer fabrication is one piece of the larger engineering effort.

Khare says he can deliver wafers to outside customers by the end of the year. For potential customers such as Rigetti the test will be how reliably those chips meet increasingly demanding designs. As Kulkarni puts it: “If we tighten the specification, what would the yield be?”

Chips & SupplyUnited States

SK Hynix’s Solidigm Is Said to Pick Banks for US IPO Next Year

Credited to Bloomberg$ · bloomberg.com2h ago

Useful

SK Hynix Inc.’s Solidigm has picked lead banks for its US initial public offering next year, according to people familiar with the matter.

View the article · $Bloomberg

Front Method

Capital · United States

Aehr Expands AI Chip Burn-In Capacity with $6M Hyperscaler Order

Credited to Semiconductor Digest · semiconductor-digest.com · 3h ago

Aehr Test Systems, a provider of test and burn-in solutions for semiconductor devices used in artificial intelligence (AI), silicon photonics, data center, automotive, and industrial applications, today announced approximately $6 million in follow-on production orders from its lead hyperscale customer for package-level burn-in (PLBI) of its next-generation AI processor for data center training and inference applications.

The orders include additional Sonoma™ high-power package-level test and burn-in systems, fully turnkey burn-in modules (BIMs) and device-specific sockets, as well as production software enhancements. The systems and consumables will be deployed at one of the customer’s outsourced semiconductor assembly and test (OSAT) partners in Taiwan. Deliveries under the order are scheduled over the next six months.

This next-generation AI processor is the second from this customer to use Aehr’s Sonoma platform for production burn-in. Following the previously announced initial production win, these orders expand the customer’s capacity ahead of an expected production ramp of this second AI device in calendar 2027. The customer has indicated that this second AI processor is expected to ultimately reach production volumes exceeding those of the customer’s first processor brought into production burn-in on the Sonoma platform. That first processor is already in high-volume production, with volumes expected to continue ramping significantly as additional Sonoma system capacity is deployed over the next several months.

“These follow-on production orders demonstrate the continued expansion of our relationship with this hyperscale customer and its confidence in Sonoma to support successive generations of increasingly powerful and complex AI processors,” said Gayn Erickson, President and Chief Executive Officer of Aehr Test Systems. “The program is progressing from qualification and initial production systems to additional systems, consumables and software supporting the anticipated production ramp. As volumes increase, we believe there is potential for significant additional demand for Sonoma systems and device-specific consumables.”

The next-generation processor requires independently controlled thermal zones within each package to achieve the operating conditions needed for effective production burn-in and early-life failure screening. Sonoma’s multi-zone thermal control capabilities enable independent temperature management across different areas of a device while delivering the high current required to electrically stress the processor.

The platform also supports independent electrical test vectors and monitoring of multiple devices or functional elements within a multi-chip package. These capabilities address the differing power densities and thermal characteristics of advanced AI packages, which increasingly combine compute die, high-bandwidth memory and interface components.

“As AI processors become more complex, controlling the temperature of the package as a whole is no longer sufficient for some applications,” Erickson continued. “Sonoma’s ability to independently manage multiple thermal zones and electrical test functions helps customers address these challenges. We believe these capabilities will become increasingly important as AI packages continue to grow in power, size, complexity and value.”

Aehr is also engaged in detailed discussions with this customer regarding reliability qualification and production burn-in requirements for future generations of higher-power, larger AI devices. These discussions include both Aehr’s Sonoma package-level and FOX-XP™ wafer -level burn-in (WLBI) platforms.

“As the value of advanced AI packages increases, identifying early-life failures before expensive packaging can deliver significant economic benefits,” Erickson added. “WLBI can screen die before assembly, while PLBI can stress and screen the completed device. Customers may choose either approach or use both at different stages of qualification and production, depending on their requirements.”

“Our ability to support both approaches positions Aehr well as customers develop reliability strategies for future AI processors. With Sonoma supporting this customer’s first processor in high-volume production and now its next-generation program, we are building a foundation for continued growth as the customer’s AI roadmap advances.”