A pivotal part of quantum information research, quantum computing requires accurate qubit operations and readout. To achieve fault-tolerant quantum computation, fast and high-fidelity readout of quantum states is essential. Some technical methods have been developed for qubit-state discrimination. However, traditional methods are either insufficient or time consuming.

Innovatively, using recently popular machine-learning (ML) tools, USTC researchers managed to design methods to guarantee both low time consumption and high-fidelity readout on ion-trap systems. The research appeared July 22 in the journal Physical Review Applied.

Called ML-assisted single-qubit readout methods, the implementation is conducted on a thoughtfully made hardware module. Considering that the field-programmable-gate-array (FPGA) based system processes data faster, researchers implement real-time state discrimination with ML algorithms on an embedded-hardware system, without using traditional CPU or GPU. In more detail, they test several methods including fully connected neural networks, convolutional neural networks (CNN), etc. They put CNN into detailed experiment and show that only 43 sub-bins (approx. 129 μs) are needed by CNN method, while threshold method, one of the traditional, needs doubled. Therefore, CNN method proves significantly reduced detection time needed for qubit discrimination.

Figure 1. The infidelity of single-qubit readout in CNN method. Image by DING Zihan et al.

In the paper, the researchers show that for single-qubit readout in a trapped-ion system, they can achieve 99.5% fidelity within around 170 μs per sample, with a FPGA- and ARM-processor-based fully connected neural network method. The proposed ML methods for single-qubit readout reduce half of the inference time in simulation, compared with traditional methods. Moreover, the hardware implementation further improves the efficiency by at least 3.

Forwardly looking, the embedded-qubit readout method could be extended to synchronous multi-ion-state readout and fast feedback control, hence promising for more flexible quantum-gate operations in the future.

(Written by QIANG Jiaxuan, edited by LI Xiaoxi, USTC News Center)