IBM has built a modular cryogenic cooling subsystem that reaches 15 millikelvins, more than 100 times colder than outer space. The two-module architecture stands more than eight feet tall and eight feet wide combined. It anchors the company's 2029 roadmap for Starling, its first fault-tolerant quantum system.

Why does cooling matter for superconducting quantum machines?

IBM, Google, Rigetti, and IQM all rely on superconducting qubits. These qubits operate only near absolute zero, between 10 and 20 millikelvins. Heat-driven atomic vibrations would otherwise tear apart the electron pairs that carry quantum states. Calculations would collapse before completion.

The new subsystem reaches 4 Kelvin, the temperature of liquid helium, in under five days. Its 15 millikelvin floor follows soon after, IBM said. Jerry Chow, IBM Fellow and chief technology officer for quantum-centric supercomputing, unveiled the system with Oliver Dial, IBM Fellow and vice president of quantum systems.

"It's a real physical architecture for fault-tolerant quantum computing and the success in being able to actually connect these modules really demonstrates the complex engineering that we need to go in terms of pushing beyond just single quantum processor," Chow told journalists.

What does Starling actually demand?

Starling targets roughly 200 logical qubits, according to IBM's stated roadmap. It will run up to 100 million quantum operations, a 20,000-fold lift over today's IBM quantum machines. Fault tolerance and error correction are central design goals.

"When we talk about scaling, it is not just about building bigger and bigger chips at the processor level," Chow said. "It is really about all the infrastructure and the supporting pieces around it as well in the system. This is our first shared ultra-cold environment that allows multiple chips within to be connected together."

The modules include doors with metal EMI gaskets that block electromagnetic interference. Rubber O-rings create vacuum seals. Layered mylar sheets act as "super insulation," according to Dial. Internal metal shields step temperatures down toward a dilution refrigerator at the base.

How does modular design address scaling?

The two modules decouple for shipping. Customers can reconnect them on site, simplifying maintenance and future upgrades. Starling will likely include about a dozen such modules, Dial said.

L-couplers form the connective tissue. These aluminum superconducting cables, first tested on the Flamingo proof-of-concept multi-chip processor in 2024, span up to about one meter. Standard on-chip couplers reach far shorter distances and limit chip placement.

"It is really critical to us because it forms a foundation of our modular designs," Dial said. "The ability to get the processors far apart gives us more space for wiring, gives us the ability to independently test and debug and replace them."

What are the next concrete steps?

Later this year, IBM will run a demonstration with one Nighthawk chip per cryogenic unit. Nighthawk is a 120-qubit QPU that IBM launched last year. By Starling's arrival, each module is expected to hold about 2,000 physical qubits, roughly four QPUs.

The new modules offer about twelve times more wiring area than IBM Quantum System 1. IBM is sourcing cryogenic engines from established vendors, including Bluefors, a Helsinki-based specialist whose dilution refrigerators already run at IBM's Poughkeepsie, New York labs. Chow said Bluefors hardware will speed engineering and simplify long-term service.

What remains unproven?

Chow framed the remaining work as engineering execution, not fundamental science. "We've nailed down the science for fault-tolerant quantum computing, and instead a big part of what we're doing to get there now is engineering," he said. "It's not about a single breakthrough to get to fault tolerance. It's really about thousands of these little engineering feats."

Starling targets 2029. IBM has not published a detailed milestone schedule between this week's announcement and that launch. The roadmap hinges on L-couplers, Nighthawk successors, error-correction advances, and the smooth integration of cryogenic modules at customer sites. No component has yet been demonstrated at Starling's planned scale, and IBM has not disclosed system pricing or Starling's total qubit count beyond the 200 logical-qubit figure.