Why Quantum Computers Run Out of Cable

The hardest part of building a quantum computer is no longer the chip. It is the wiring around it.

IBM connected the first two modules of a new cryogenic architecture this week at its Poughkeepsie, New York facility, calling it a milestone on the path to a fault-tolerant machine by 2029. Each module cools to below 15 millikelvins, just a hair above absolute zero, and offers up to 12 times more wiring space than the company's most widely used quantum systems. The box-shaped design puts the two modules in a shared cryogenic environment with considerably more room for chip-to-chip interconnects than earlier IBM hardware.

No quantum processors are running across the two modules yet. IBM plans to install one Nighthawk processor, a next-generation superconducting chip, in each module later this year and demonstrate a processor-to-processor connection between them. The milestone, in other words, is operational cryogenic hardware, not yet a working multi-processor system.

A qubit is the basic unit of quantum information, and running a superconducting one takes temperatures near absolute zero. Between the room-temperature control electronics and the chip sit chains of equipment: coaxial cables, attenuators that step signal levels down as they approach the cold stage, amplifiers, couplers, and connectors. Today's systems need roughly 2.5 cables per qubit. Janne Lehtinen, the CTO and co-founder of SemiQon, a competing cryogenic components vendor, calls that density the real constraint. Millions of connectors in current setups, he told EE Times, and the mechanical complexity of assembling them, are now the reliability bottleneck.

Brute-force extension of that architecture runs out of room somewhere in the thousands to tens of thousands of qubits, Lehtinen said. The picture is not a single dramatic wall but a slow crush of signal-chain parts. Each cable is a candidate failure point, and each connector is a place where thermal contraction, vibration, and human assembly error can degrade performance. A 100,000-qubit machine following the same recipe needs roughly 250,000 cables and an order of magnitude more connectors, all of them carrying signals that must survive the trip from room temperature to near zero without losing fidelity.

IBM's new system is a direct response: more physical room for wiring, more room for chip-to-chip connections both inside and between modules, and a layout meant to host multiple processors in one cryogenic environment rather than treating each as its own self-contained fridge. At the temperatures a superconducting processor needs, every wire that crosses between stages has to be carefully engineered to carry signal without leaking heat, and the same goes for every connector that joins one wire to the next. More space means more paths for those wires and connectors, and more flexibility in how chips inside a module can be linked to chips in its neighbor. The Nighthawk chip, once installed, is the first concrete thing the architecture will be asked to host.

The 2029 target, a fault-tolerant quantum computer, is IBM-stated and not independently verified. Fault-tolerant here means a machine whose logical qubits, error-corrected groupings of physical qubits, can run long algorithms without noise overwhelming the computation. Reaching it still depends on error-correction overhead, which can demand many times more physical qubits per useful logical qubit, plus qubit quality and software stacks the new modules do not address. What the modules do address is the build-up of cables, connectors, and signal-chain parts that scale with qubit count, and the fact that this is now where scaling is decided. A system that needs roughly 2.5 cables per qubit needs roughly 50 million cables for a 20-million-qubit machine, and every connector along the way is a place where the build can fail. IBM's design adds room. It does not yet prove that room is enough.

The next checkpoints are concrete and easy to track. Nighthawk processors go into each module later this year. A working processor-to-processor link across the two modules follows. From there the metric to watch is cabling density per logical qubit, not raw qubit count, and the next bottleneck on the list is the one IBM has not yet addressed: control electronics, signal integrity at scale, and the cost of running millions of cryogenic interconnects in a production environment. IBM has put a year on its fault-tolerant target but not on the first end-to-end test of Nighthawk across both modules. The cabling math behind both dates is the watch item from here.