Quantum error‑correction progress

Quantum computers still make mistakes so often that useful programs usually fall apart before they finish. Two new April 2026 papers focus on the same fix: spreading one “logical” qubit across many physical qubits so single errors can be found and corrected. (arxiv.org 1) (arxiv.org 2) One paper, posted to arXiv on April 18, lays out error-correcting codes for reconfigurable neutral-atom arrays, where laser “tweezers” trap atoms and move them into new patterns during a computation. The authors from QuEra, Harvard, and the Massachusetts Institute of Technology say their design can keep encoding rates above 1/2, meaning more than half the qubits in a block carry logical data instead of overhead. (arxiv.org) (quera.com) Under a circuit-level noise model with a physical error rate of 0.1%, that neutral-atom paper reports per-logical-per-round error rates of about 1.3×10^-13 for a [[2304,1156,≤14]] code and about 2.9×10^-11 for a [[1152,580,≤12]] code. The paper says those numbers approach the “teraquop” regime, shorthand for running roughly a trillion logical operations with about one failure. (arxiv.org 1) (arxiv.org 2) A second paper, submitted April 21 by IonQ researchers, does not claim a new record error rate. It describes a full trapped-ion machine design called “Walking Cat,” including the compiler, the error-correction routines, the chip-level layout, and a decoder fast enough to keep up with the hardware. (arxiv.org) (ionq.com) In trapped-ion systems, qubits are stored in charged atoms held in electromagnetic traps, and the ions can be physically shuttled between zones for gates, measurement, and reset. IonQ says its architecture uses “cat states” as probe states for fault-tolerant measurements and a quantum charge-coupled-device layout to move ions around the chip. (ionq.com) (arxiv.org) The IonQ paper’s dense design encodes 110 logical qubits and, in simulation, reaches about 1 million T gates per day using 2,514 physical qubits. The same paper estimates that a 10,000-physical-qubit version could simulate a 100-site Heisenberg model within one month, including the shots needed for chemical-accuracy targets. (arxiv.org) Both papers are trying to cut the same bottleneck: the huge qubit tax imposed by error correction. A 2025 Nature Communications paper on yoked surface codes said conventional surface-code approaches can require 1,000 to 2,000 physical qubits per logical qubit at error rates around 10^-3 for algorithmically relevant workloads. (nature.com) The neutral-atom result is a code-and-decoder study matched to hardware that can rearrange atoms quickly, not a report that a 2,304-qubit logical memory has already been run in the lab. The trapped-ion result is a blueprint built from components the authors say have been demonstrated on small devices, not a finished fault-tolerant computer. (arxiv.org 1) (arxiv.org 2) That leaves the field with two different bets on the same problem. One uses movable neutral atoms to pack logical information more densely; the other uses shuttled ions and cat-state measurements to spell out how a fault-tolerant machine could actually be organized. (arxiv.org) (arxiv.org)