Stability experiment

The stability experiment is the dual of the standard QEC memory experiment, introduced by Craig Gidney (Google Quantum AI) in a 2022 paper (Quantum 6, 786). Where a memory experiment checks that a logical observable survives across time, a stability experiment checks that the error-correction system (syndrome-extraction circuits, measurement, and the classical decoder acting together) can correctly determine the product of a large region of stabilizers across space. That operation underlies moving logical qubits and lattice surgery, so the benchmark certifies a capability fault-tolerant computation needs but memory experiments never exercise. Though the patches are modest, what it certifies is collective: combining many stabilizer measurements into one correct product, not any individual gate or qubit.

How it works

Run a patch whose boundary conditions make the product of a region of stabilizers a known, fixed global invariant. Measure syndromes for a varying number of rounds, decode, and check whether the decoder reconstructs the invariant correctly; the score is the logical error rate of that decoded product versus round count. Space and time swap roles relative to a memory experiment: adding rounds suppresses the failure probability (rounds act as the effective code distance) while patch size plays the role that duration plays in a memory experiment.

Strengths and limitations

It probes timelike failure mechanisms (measurement and decoding errors) that memory experiments under-weight, is cheap to run, and scores the full QEC stack including real-time decoding. Because the roles of space and time are exchanged, its numbers are not directly comparable to a memory experiment’s logical error per round. Adoption is growing but still partial (flagship results such as Google’s below-threshold paper report only memory experiments), and the original paper validated the protocol in Stim simulation, with hardware demonstrations following from 2024. Not to be confused with “stability” as device-parameter drift in NISQ benchmarking.

Notable results

Riverlane and Rigetti ran an 8-qubit stability experiment with up to 25 decoding rounds to validate a real-time FPGA decoder (arXiv:2410.05202; Nature Communications, June 2026). Reference circuits ship with Stim’s circuit data references, and the protocol appears across recent surface-code, color-code, and lattice-surgery literature.