IBM links two 120-qubit chips with four superconducting quantum cables
IBM researchers report a four-channel link between two 120-qubit processors, with 99% average state-transfer fidelity in an arXiv preprint.
IBM researchers say they have demonstrated a four-channel quantum interconnect between two 120-qubit superconducting processors, a step toward modular systems built from multiple chips rather than a single larger die. In a study posted to arXiv on October 27, the team reported average quantum state-transfer fidelity of 99% across meter-scale superconducting links, along with cross-chip gates averaging 94.75% fidelity. The work is a preprint.
What was announced
In the preprint, IBM connected two Nighthawk R1 processors, each with 120 qubits, using four aluminum coaxial cables roughly 80 centimeters long inside a cryogenic system. The researchers used the links to transfer quantum states, perform operations between qubits on separate chips, and generate entanglement spanning both processors.
The company describes the result as the first demonstration, to its knowledge, of a multichannel quantum interconnect between large superconducting quantum processors. The setup did not involve all 240 qubits. Instead, the team wired and measured selected qubits near the interconnects to speed development.
The technical details
Each link used dedicated qubits at the ends of the cable, with tunable coupling elements controlling interactions between those qubits and the microwave link. Using network benchmarking, the researchers measured state-transfer fidelity in both directions and reported that all four links exceeded 98.86%. The best link reached 99.29%, with an uncertainty of 0.07 percentage points.
The team then used the connections to implement controlled-Z gates between computational qubits on different processors. Those remote gates required two state transfers, three local two-qubit gates and additional single-qubit controls. Average fidelity for these distant gates was 94.75%, while the best reached 96.42%, with an uncertainty of 0.28 percentage points.
IBM also reported entangled states distributed across the two chips. The researchers generated two entangled pairs simultaneously through adjacent links, although one pair showed lower fidelity during simultaneous operation. They further created four-qubit and six-qubit entangled states spanning the processors. Measured fidelities were 77.23% for the four-qubit state and 67.67% for the six-qubit state; after statistical correction for measurement errors, those values rose to 91.70% and 90.87%.
Packaging and reliability
A notable part of the work was mechanical and cryogenic integration. The cables attached perpendicular to the chip surface, a geometry the researchers say is better suited to denser interconnect layouts than earlier parallel arrangements. The attachment used indium-to-indium pressure bonds formed at room temperature after the processor modules were mounted in the cryostat.
According to the study, all four links stayed operational through three warm-up and cool-down cycles without connection adjustments.
The paper also identifies magnetic fields during cooldown as a source of cable loss. IBM added two layers of magnetic shielding around the cables, though the researchers describe the shielding braid as a temporary measure.
Why it matters
The result addresses a central scaling question for superconducting quantum computing: whether multiple processors can be linked without losing too much of the quantum information needed for useful computation. Building larger systems from modules could ease physical constraints around chip size, wiring and packaging.
At the same time, the data show the remaining gap between basic state transfer and full distributed operations. IBM’s error model attributed about 77% of the state-transfer error to limited coherence in the cable and dedicated qubits, especially dephasing. The researchers say they intentionally skipped some fabrication and processing steps known to improve qubit performance in order to accelerate the experiment.
What to watch
The next test will be whether IBM can turn four working links into a larger number of reproducible, manufacturable connections while raising fidelities. The paper also points to a possible architectural change: integrating interconnect elements directly into the processor lattice to reduce the number of steps required for remote gates.
For now, the work demonstrates a promising subsystem rather than a complete modular quantum computer. As the authors put it, the challenge ahead is reliable, repeated use of such links in error-corrected computation.
Sources
Facts on this page come from the sources below, retrieved by DesignQuantum.