IBM links two separate quantum chips into one machine with cable-based 'L-couplers'

Researchers at IBM have connected two separate superconducting quantum processors into a single working system using four cable-based links, reporting an average quantum-state transfer fidelity of 99.00 percent between the chips. The result, posted as a preprint on October 21, 2026, and described in an IBM Research blog post on October 27, is an early demonstration of "modular" quantum computing: building a larger machine by wiring together multiple chips rather than trying to pack more qubits onto a single piece of silicon.

The two processors are IBM Quantum Nighthawk R1 chips, each carrying 120 qubits. IBM's team, led by physicist Neereja M. Sundaresan and including 41 co-authors, joined them with four interconnects the company calls "L-couplers," installed inside one dilution refrigerator alongside the chips.

How the link works

Each L-coupler is built around roughly 80 centimeters of superconducting aluminum coaxial cable, paired with simplified tunable qubits ("L-qubits") and RF-SQUID-based couplers at each end that shuttle a quantum state onto the cable and back off again at the other chip. IBM attaches the cables perpendicular to the chip surface rather than running them parallel to it, a geometry the company says leaves room for denser wiring in future, larger systems. The physical connections themselves use room-temperature indium-to-indium pressure bonds, made after the two processor modules were already mounted in the cryostat — and they held up through three full warm-up-and-cool-down cycles without needing to be redone.

Across the four links, the team measured an average state-transfer fidelity of 99.00 percent, with every link reaching at least 98.86 percent and the best hitting 99.29 percent. A more demanding test — a two-qubit "CZ" gate performed between qubits on the two different chips, rather than a simple state hand-off — averaged 94.75 percent fidelity, with the best link reaching 96.42 percent, The Quantum Insider reported. The team also generated entangled pairs of qubits across the two processors, including two pairs simultaneously through adjacent couplers, and larger four- and six-qubit entangled states that reached 77 and 68 percent fidelity before error correction and roughly 92 and 91 percent after it.

The resonator modes inside the couplers showed an internal quality factor of about 940,000, a measure of how little energy the cable-based link loses, which IBM's team considers workable for carrying quantum information over that distance. The main source of error, by the team's own accounting, was dephasing in the L-qubits, which were built without IBM's usual coherence-optimizing fabrication steps in order to move the project faster. Stray magnetic fields trapped in the metal during cooldown also degraded the cables' performance; the team added two layers of magnetic shielding as a temporary fix rather than a permanent solution.

Why chip-to-chip links matter

Superconducting quantum processors face a practical ceiling on how many qubits can be crammed onto one chip: wiring space, fabrication defects and chip real estate all get harder to manage as a single piece of silicon grows. The alternative pursued here is to keep individual chips at a manageable size and link several of them together so they function as one larger processor — the same logic that pushes classical computing toward multi-chip servers rather than ever-larger single processors.

IBM's publicly posted quantum roadmap ties this approach directly to its stated goal of a large-scale, fault-tolerant quantum computer called Starling by 2029, which the roadmap says is meant to run on the order of 100 million quantum operations across 200 error-corrected logical qubits — a scale that would require combining many physical qubits across multiple modules. The roadmap also notes that IBM's current commercial systems, the 120-qubit Nighthawk and 133- to 156-qubit Heron processors, are already deployed at IBM Quantum System Two installations in New York, Kobe and San Sebastián, with a Chicago site underway; L-couplers are presented as the connective layer meant to eventually link such systems together rather than requiring one monolithic chip.

Caveats and what is not yet shown

The 99 percent figure describes moving a quantum state from one chip to another, not running a full quantum algorithm across both. The weaker two-qubit gate fidelity (94.75 percent) and the still-lower multi-qubit entangled-state fidelities indicate the link is currently better at simple transfers than at the more complex operations an error-corrected computer would need to perform between chips. The findings have not yet been through peer review; they currently exist only as an arXiv preprint. IBM's own modeling suggests that improving the L-qubits' coherence time to about 100 microseconds — not yet achieved — could push transfer error below 0.1 percent and distant two-qubit gate fidelity above 99 percent, but that projection is a model, not a measured result.

What comes next

In its blog post, IBM said its near-term priorities are improving the L-couplers' manufacturing yield and reproducibility, reducing their error rates, and scaling up to systems where many more couplers connect and, eventually, disconnect larger and more complex processors, Network World reported. The company said it intends to build out connected Nighthawk-based systems using this approach as a step toward the hardware that would eventually support quantum error correction and the planned Starling machine. No specific date was given for when a larger multi-chip system using L-couplers might be deployed.