The Signal
USTC Researchers Demonstrate Four-Way Multiplexed Quantum Gates
A paper by researchers affiliated with the University of Science and Technology of China reports parallel control of four single-qubit and four two-qubit gates using frequency-multiplexed signals. The experiment found no statistically signi
Quantum.world Team · Written with AI · fact-checked ·
A paper posted to arXiv on October 26 reports a four-way frequency-multiplexed demonstration of single- and two-qubit gates on a superconducting processor. The authors include researchers affiliated with the University of Science and Technology of China. The paper says the approach could cut the number of control lines by one to two orders of magnitude. That larger reduction is a projection: the experiment multiplexed four gates, while the more ambitious figures come from a modeled architecture.
Four gates share microwave control
Frequency-division multiplexing assigns separate microwave frequencies to signals sharing a control line. The researchers combine gate pulses into a multi-frequency waveform, then use band-pass filters to route each frequency to its target qubit or coupler. Unlike time-division multiplexing, which sends operations in separate time slots, this method allows four selected gates to run in parallel.
The work addresses a gap in superconducting-qubit control. Readout and some single-qubit operations already use multiplexed signals, but two-qubit gates have commonly relied on unipolar flux pulses without a carrier frequency to distinguish simultaneous operations. The team instead used microwave pulses to drive controlled-Z gates through adjustable couplers, alongside frequency-multiplexed single-qubit rotations.
“These results show that frequency sharing can be introduced at the gate level without a measurable fidelity penalty, providing the missing gate-control element for an all-microwave architecture supporting universal quantum control.”
Dayu Li and his co-authors write in the paper.
Pre-distortion offsets filter effects
In the four-way experiment, the average reported error was 10.57 × 10-4 for parallel single-qubit XY(π/2) gates and 6.43 × 10-3 for parallel controlled-Z gates. The researchers compared those results with dedicated-channel baselines and found no statistically significant increase in error from multiplexing at the experiment’s measurement resolution.
The filters distorted the control pulses and initially raised measured errors. The team applied waveform pre-distortion to compensate for the filters’ response; for single-qubit gates, it also accounted for the power splitter used to create the shared signal path. The result is evidence that the added filtering and signal combination can work in this setup, not a demonstration that every larger processor will retain the same gate performance.
The larger wiring cuts are projections
The proposed architecture combines frequency-division multiplexing with time-division multiplexing and on-chip DC sources. In the authors’ periodically repeated arrangement, four coupler groups take turns using shared control resources, while individual couplers within an active group receive distinct microwave frequencies. Under the gate parameters used in the experiment, the modeled arrangement reduces the control-line count to one-twelfth of a dedicated-line baseline.
A more aggressive one-forty-eighth ratio applies only under the paper’s stated idealized assumptions. The projection uses a repeated unit of 32 qubits and 64 couplers; the physical experiment demonstrated four-way multiplexing, not a processor at that scale. Frequency spacing, crosstalk, available control bandwidth and gate performance all shape how many signals the architecture can share.
Cooling estimate leaves out other heat loads
The paper also estimates the passive cable heat load at the refrigerator’s mixing-chamber stage for a processor with 381,000 physical qubits. Using a cited reference value of 350 picowatts per cable channel, the authors calculate about 400.05 microwatts for the dedicated-line baseline and 33.34 microwatts for the one-twelfth case. The idealized one-forty-eighth case falls to 8.33 microwatts.
Those figures cover the passive cable contribution, not the refrigerator’s total heat budget; the paper explicitly excludes other cryogenic loads. The authors estimate that the control-cable contribution would fall below a representative 30-microwatt cooling level at a multiplexing ratio of roughly 13.3. The experiment establishes four-way parallel gate control; reaching the larger projected savings will depend on extending that arrangement while preserving frequency separation and gate fidelity.