27 stories
Show: Summaries / Headlines only
A learning agent used error-detection signals from surface and color codes on Willow to adjust more than 1,000 control settings. The team showed steering within a single long computation only in numerical simulation.
IonQ's team ran qLDPC, toric, and concatenated codes on one chain of 40 barium ions without moving them. In the best code, logical qubits lasted 3.95 seconds, against 3.84 seconds for the physical qubits in the authors' revised analysis.
The team measured fewer logical errors in the larger of two codes, built from 64 physical qubits, over up to eight rounds. Over longer runs, the authors concluded only that logical error rates did not grow with code distance.
Teams publishing in Science measured quantum learning advantages on quantum data, including an 11.8-order cut in samples. On ordinary datasets, classical models led on four of five in a peer-reviewed benchmark. HSBC and IBM reported gains in a preprint.
Silicon spin qubits now come from 300 mm lines at imec, Intel, and STMicroelectronics. No maker has yet shown a fab supplying enriched silicon, qubits uniform enough to tune automatically at scale, or control electronics within a million-qubit power budget.
A Shenzhen team encoded two logical qubits in four phosphorus nuclear spins with an error-detecting code. Its logical gates averaged about 86% fidelity, below the more than 95% its physical gates averaged.
In PNAS, Oxford's Tim Palmer proposed that quantum computers lose their advantage above about 1,000 perfect qubits. He wrote that his theory and standard quantum mechanics agree on small systems. No experiment has yet tested where they differ.
