The Signal
IBM Team Entangles Logical Qubits on Nighthawk Processor
Researchers report entangling two logical qubits on a 120-qubit superconducting processor, with Bell-state fidelity above 93%. Their experiments also show how excluding underperforming components improved a distance-5 surface code and how f
Quantum.world Team · Written with AI · fact-checked ·
A larger quantum error-correction code should suppress errors, but IBM’s Nighthawk revision 2 processor did not make that improvement automatic: underperforming components weakened the larger code. In a preprint posted October 23, University of Sydney and IBM researchers report that excluding troublesome components helped a distance-5 surface code cut its logical error rate from 1.9% to 0.9% per round. The team also entangled two logical qubits with Bell-state fidelity above 93%, using a technique called lattice surgery.
Nighthawk’s larger code meets uneven hardware
The experiment ran on ibm_phoenix, a square-lattice processor with 120 coupled superconducting qubits and fast hardware reset. The researchers encoded information in distance-3 and distance-5 surface codes, which use different numbers of physical qubits to protect a logical qubit against errors. A larger code can correct more errors in principle, but the authors found that variations in component performance limited that advantage on this device.
They responded by adapting the measurement circuits to avoid weak components, a method the paper calls ancilla re-purposing. In the distance-5 memory experiment, that approach brought the logical error per round below 1% and reduced it from 1.9% to 0.9%. The authors say the adjusted distance-5 code performed comparably to the best distance-3 codes occupying the same chip area—not a blanket win for larger codes, but evidence that the hardware’s weaker spots can matter as much as code size.
Lattice surgery joins two encoded qubits
To entangle logical qubits, the team used lattice surgery: it temporarily joins neighboring surface-code patches by measuring operators along their shared boundary, then separates them again. The joint measurement creates a logical Bell state, linking two encoded qubits rather than just two individual physical qubits. The reported fidelity exceeded 93%, based on measurements of the logical XX and ZZ observables.
That result has limits the paper spells out. The team post-selected for runs in which both logical qubits were correctly initialized before surgery; those accepted runs represented a yield of about 40%. Because readout error in the logical Y basis was high, the researchers did not use a YY measurement and instead bounded the Bell-state fidelity with XX and ZZ measurements.
Three logical qubits enable teleportation
The researchers fitted three distance-3 logical qubits onto the 120-qubit processor by compressing the region used for lattice surgery. They used the arrangement to teleport a range of logical states and gates, reporting fidelities from 78% to 93%. They also transferred a Bell pair between two logical qubits that were not directly coupled, with fidelity bounded between 85.4% and 91.4%; only 1.2% of runs survived the required initial-state post-selection.
The experiments demonstrate logical Clifford operations, but not a complete universal fault-tolerant gate set. The authors prepared logical T states without fault tolerance, and they say more logical qubits and substantially better fidelity are needed before quantum computers can solve industry-relevant problems.
“However, we must progress further before we can produce a quantum computer capable of solving industry-relevant problems.”
Robin Harper and co-authors, authors of the paper, write.
Fast reset slows leakage’s accumulation
Fast reset also helped in a separate stability experiment, which tested repeated rounds of stabilizer measurements. Leakage occurs when a superconducting qubit leaves the two-state space used to encode information; if it accumulates, error rates can stop falling as a circuit runs for more rounds. The team used a “walking” circuit that alternated data and ancilla roles, so each qubit in the test patch was reset every two rounds.
Across 50 million runs of a 170-round experiment, the researchers recorded eight failed decodings, putting the observed error probability near 10-7. That long-run stability test is distinct from the logical-qubit operations, which used fast measurement and reset but not the walking circuit. Together, the results show both what Nighthawk enabled in this experiment and the gap the authors identify: stronger logical operations will require higher fidelity, more logical qubits and support for non-Clifford gates.