Cosmic rays hit quantum computers. A high-energy particle strikes the superconducting chip, deposits energy in the substrate, breaks Cooper pairs, and generates quasiparticles that diffuse across the chip. Multiple qubits relax simultaneously — a correlated error burst that is far more damaging than independent single-qubit errors. This much was known. Error correction codes are designed to handle it. The standard model of radiation damage is: particle arrives, energy deposits, qubits relax (transition from excited to ground state). The error is energy loss.
Yelton et al. (arXiv:2603.13124, March 2026) fire controlled electron pulses from a linear accelerator into a dilution refrigerator holding a multi-qubit chip. By controlling the energy, timing, and position of the radiation, they can dissect the error anatomy with precision that cosmic-ray studies cannot match.
The damage is broader than relaxation. The accelerator reveals qubit excitation errors (ground-to-excited transitions, not just excited-to-ground) and frequency detuning errors (the qubit's operating frequency shifts temporarily). These error types are harder to detect in cosmic-ray studies because they're rare and intermixed with dominant relaxation events. With a controlled source, they're cleanly separable.
The error signature depends on junction geometry. Qubits with different Josephson junction designs show different error profiles under the same radiation. The physical structure of the qubit determines not just its sensitivity to radiation but which types of errors radiation produces.
This matters for error correction. Standard codes assume a specific error model — predominantly relaxation. If radiation also produces excitation and detuning with non-negligible probability, the codes face a broader threat than their design assumes. The protection is tuned to the wrong error. And the wrong error varies with hardware design, meaning the error correction strategy must be co-designed with the physical qubit, not layered on top.