Flux tunable superconducting quantum interference devices (SQUID) are widely applied in today superconducting qubit architectures, e.g., flux qubit, tunable transmon, tunable coupler. The flux control requires sending flux pulses with desired waveform to bias the flux inside the SQUID loop. But many devices and imperfections can distort the waveform, inducing undesired flux fluctuations. Such waveform distortions will cause inaccurate frequency biasing or frequency damping, which leads to phase errors accumulation of a qubit (or coupler) and decrease the fidelity of qubit Z-control (e.g., iSWAP gate and CZ gate).
The distortions comes from many sources: analog filters (e.g., low-pass filter, high-pass filter), Bias-Tee, AWG, junction points on the transmission line (causing transmission and reflection of the EM wave), skin effect, on-chip resonance, etc.
We aim at exploring new techniques of combining reinforcement learning with digital inverse filtering methods (inverse IIR, and FIR filters), to effective and fast calibrate and fix the distortions. The predistortion method uses inversed digital filters, calibrated by the transient analysis of each components contributing distortions, to predistort the flux pulse inside the AWG, and all distortions can be countered.
This technique will be widely applied in future, and we are currently developing the inverse filter calibration code and the reinforcement learning model.
Jiheng Duan 段繼恆
First year PhD student
My research interests include superconducting quantum computing, high fidelity two-qubit gate, and distortion correction of digital signals.