Demonstrating bosonic control beyond the ancilla limit

Superconducting bosonic systems offer a promising platform for exploring continuous variable quantum physics and hardware-efficient error correction, with even single-oscillator systems competing with industry-scale processors containing dozens of qubits [1,2]. However, the performance of these systems is limited by two key aspects. First, while an isolated linear oscillator offers a large Hilbert space and a simple error structure (dominated by photon loss), controlling this oscillator necessarily introduces non-linear effects that reduce the viable Hilbert space size and introduce new errors. In fact, state-of-the-art demonstrations of bosonic gate fidelities and error correction gain are primarily limited by ancilla errors, regardless of the encoding choice. Second, achieving practically useful error rates necessarily requires concatenation with a many-body logical code, which requires challenging high-fidelity multi-mode control.

We address these two fundamental issues by introducing a modular scalable bosonic architecture based on `linear’ parametric couplers (LINCs). The LINC has a symmetry-engineered Hamiltonian that makes it close to linear while idle, but able to turn on clean parametric Gaussian operations when driven. We use the LINC to construct universal control and tomography chips, and place these chips in modules connecting high-Q on-chip bosonic memories. The LINC separates the bosonic mode from the ancilla, with idle dispersive shifts suppressed to less than 300 Hz, and oscillator dephasing times reaching 20 ms. We show that this separation allows multi-oscillator Gaussian operations without involving the ancilla, and both single-qubit gates and reset on the ancilla with minimal back-action on the oscillators. When interaction is desired, we dynamically activate dispersive shifts by four orders of magnitude in < 50 ns, using it for dispersive control, error-syndrome measurement, and full Wigner tomography. Careful modulation of this parametric dispersive interaction even allows pulses tolerant to ancilla decay and dephasing. Together, this architecture scalably demonstrates bosonic control and error-detection beyond the ancilla limit.

[1] V. Sivak et al, Nature 2023

[2] Google Quantum AI, Nature 2025