Category: Error correction and logical qubits

A 280-atom machine scored up to an order of magnitude above unprotected qubits on test circuits. Commercial jobs need far more qubits.

Google’s 49-qubit code erred 2.914% per cycle versus 3.028% for 17-qubit codes. Scaling helps, but hardware must still improve by at least 20%.

Harvard and QuEra hit 99.5% accuracy on up to 60 atoms at once, past the 99% error correction needs. A planning signal, not a usable machine yet.

Three basic operations on one Sandia ion stayed below the proven 6.7×10⁻⁴ error limit. Buyers can ask for certified error rates; linking qubits is next.

Fast lookups in 8GB need quadrillions of qubits; a million qubits means about 3 years per lookup. Plans should not assume cheap quantum memory.

Each size step cut errors 2.14-fold, up to 101 qubits. It only stores data so far, but plans for useful machines now rest on a measurement.