Cosmic-Ray Correlated Errors
High-energy particles striking the chip deposit energy that spreads across many qubits at once, causing correlated errors that ordinary error correction cannot absorb.
A New Kind of Error
Most qubit errors are local and independent: one qubit fails without affecting its neighbors. Quantum error-correcting codes are built on exactly this assumption. Cosmic-ray and background-radiation events break it. When an energetic muon, gamma photon, or other particle strikes the substrate, it deposits energy that turns into a flood of phonons and quasiparticles spreading millimeters across the chip, degrading many qubits within a short window. The errors are correlated in space and time.
Why Correlation Is Dangerous
A code such as the surface code corrects a limited number of independent errors per cycle. A single cosmic-ray event can flip a large fraction of physical qubits simultaneously, far exceeding what the code can handle, and can wipe out the logical information in one stroke. Because these events are rare but catastrophic, they set a hard ceiling on the logical error rate that cannot be lowered simply by adding more physical qubits in the usual way.
- A single impact affects many qubits at once, violating the independence assumption.
- Energy spreads as phonons and quasiparticles over millimeter scales.
- Events are infrequent but can be individually fatal to a logical qubit.
Evidence
Dedicated experiments have observed simultaneous lifetime drops across arrays of qubits, timed and shaped consistently with particle impacts, and matching the known rates of cosmic-ray muons and ambient radioactivity at the surface. The recovery timescale, tens of microseconds to milliseconds, matches phonon and quasiparticle dynamics rather than any control error.
Mitigation Strategies
Approaches include operating underground or behind shielding to cut the muon and gamma flux, engineering the substrate to confine or drain the phonon energy quickly with normal-metal absorbers and gap-engineered barriers, and detecting events in real time so the affected region can be flagged and its data discarded or specially handled. Fault-tolerant architectures are also being designed to survive occasional bursts of correlated error. This is an active research frontier because it touches the ultimate scalability of error correction.