The Signal
OQTOPUS Adds Buffer-Qubit Reservations Against Crosstalk
A paper describes adding buffer-qubit reservations to OQTOPUS, a software stack for running quantum circuits from multiple users on one chip. In a test on one superconducting quantum computer, observable dephasing changes appeared only at q
Quantum.world Team · Written with AI · fact-checked ·
“Experiments on a real device characterize the attack effect in terms of energy relaxation and dephasing, showing that observable degradation appears in dephasing characteristics of qubits adjacent to the attacker circuit.”
Haruki Nakayama and co-authors, authors of the paper
The authors of a paper submitted to arXiv on October 21 describe adding buffer-qubit reservations to OQTOPUS, an open-source software stack used to operate real quantum computers. The method targets a risk in quantum multi-programming, or QMP, where a cloud service runs circuits from different users at the same time on one chip. A user’s circuit can create crosstalk that interferes with another user’s computation.
Rather than trying to identify malicious jobs, the allocation method leaves physical qubits idle between circuits. The authors say the approach is implemented in OQTOPUS and released as open-source software. Their tests establish that the allocator reserves the buffers; they do not establish that the method prevents degradation in a victim circuit’s final output.
Measurements point to neighboring qubits
The researchers measured the effects of an attack on a superconducting quantum computer at The University of Osaka, using a qubit chip provided by RIKEN. Their test circuit used repeated CNOT gates, while they measured energy relaxation, or T1, and dephasing, or T2*, in qubits assigned to other circuits. A reference run omitted the attacker circuit but added a delay to match the elapsed time.
For each data point, the team collected 10,000 measurement shots and repeated the measurement independently seven times. They compared attack and reference estimates using approximate 95% confidence intervals. The intervals for T1 included zero across the qubits tested, while observable changes in T2* appeared only at positions adjacent to the attacker circuit.
One example was qubit q2, next to the attacker circuit: for a prepared |+⟩ state, its fitted T2* fell from 2.28 ± 0.07 microseconds in the reference run to 1.95 ± 0.05 microseconds under attack. For the |i⟩ state, the paper reports observable changes at q2 and q3, also adjacent to the attacker. The measurements support reserving neighboring qubits in this experiment; they do not show that crosstalk has a one-qubit range on every device or under every attack.
OQTOPUS marks affected qubits as unavailable
The allocator maps each user circuit onto the device’s qubit-connectivity graph, then marks unused qubits within the measured crosstalk range as buffers before placing the next circuit. In the paper’s formulation, those reserved qubits cannot be used by another circuit in the same allocation batch. The authors implement the added constraints as a constraint-programming problem using Google’s OR-Tools.
To test the allocation procedure, they processed 100 randomly generated circuits, each using two to five qubits, on a 64-qubit square-lattice topology. The software placed all circuits across 12 execution batches without violating the buffer constraints. In that test, buffers accounted for 53.1% of the chip’s qubits.
The protection cost is still unmeasured
That 53.1% figure describes this particular test, not a general overhead estimate. The authors say the buffer share may fall when user circuits occupy more qubits, but they have not yet quantified the resulting throughput cost across different circuit sizes. Reserving hardware can reduce how much work a service schedules at once, and the paper leaves that trade-off open.
The team also lists end-to-end testing of victim-circuit outputs and attacks with longer-range effects as future work. For now, the result is a software-level allocation change grounded in measurements on one real device: OQTOPUS can keep adjacent physical qubits idle between co-scheduled circuits, while the effect on service capacity and computational outcomes remains unquantified.