Swedish firm FirstQFM has developed an AI model that reduces the execution time of single-qubit operations by 33% without sacrificing accuracy. The technology was tested on a 21-qubit superconducting chip, Contralto-D21, at the Israel Quantum Computing Center.
The AI model optimizes control pulses by analyzing calibration data on classical computers, eliminating the need for additional measurements on the quantum processor. The system was built using Nvidia's Leonardo supercomputer and the CUDA-Q platform.
For two-qubit operations, the model utilized existing data and brief hardware checks to improve Bell state fidelity for five qubit pairs. The results were verified through randomized benchmarking, though independent confirmation is still pending.
FirstQFM plans to expand its optimization suite for the 21-qubit processor and prepare the technology for larger QuantWare systems based on the VIO architecture. The company has not disclosed absolute values for operation duration or accuracy.
