USTC makes progress in machine learning to improve superconducting quantum bit reading efficiency

According to news from USTC, the team led by Academician Guo Guangcan of USTC has made important progress in improving the efficiency of superconducting quantum bit reading through machine learning. The research group of Professor Guo Guoping of the team cooperated with Benyuan Quantum Computing Company to study the impact of crosstalk on the reading of quantum bit states on the Benyuan "Kuafu" 6-bit superconducting quantum chip, and innovatively proposed to use shallow neural networks to identify and read the state information of quantum bits, thereby greatly suppressing the impact of crosstalk and further improving the fidelity of multi-bit reading.

It is understood that in recent years, the world has achieved high-fidelity single-bit single-shot reading and multi-bit multiplexed single-shot reading; however, due to the existence of various forms of stray coupling, the state of adjacent bits may affect the measurement results of the target bit, thereby reducing the measurement fidelity and, in turn, the success rate of the quantum algorithm. With the further expansion of quantum chips, in order to further improve the reading fidelity, how to solve the above crosstalk problem will become a major challenge facing researchers.

Figure 1. Traditional quantum bit reading scheme and the impact of crosstalk

In order to solve the problem of read crosstalk, other international research groups have previously focused on how to suppress crosstalk from a hardware level, such as configuring a separate read filter for each quantum bit's read cavity, or increasing the spatial and frequency domain distances between read cavities; although these solutions have suppressed crosstalk to a certain extent, they have had an adverse effect on the expansion and integration of quantum chips.

Figure 2 Structure diagram of the first generation "Kuafu" 6-bit superconducting quantum chip

Based on the above, Professor Guo Guoping's research group cooperated with Benyuan Quantum Computing Company to propose a new quantum bit reading scheme by abstracting and simulating the quantum bit information extraction process: by training a shallow neural network built based on the digital signal processing process, accurate identification and classification of the quantum bit state can be achieved.

The researchers applied this scheme to the original "Kuafu" 6-bit superconducting quantum chip. The experiment found that the new reading scheme not only effectively improved the reading fidelity of 6 bits, but also greatly suppressed the reading crosstalk effect; at the same time, since the data processing in the new scheme can be further simplified to a single-step matrix operation, it can be directly transferred to the FPGA in the future, thereby realizing zero-delay judgment of the quantum bit state and real-time feedback control of the quantum bit.

This solution is not only applicable to superconducting quantum computing, but also to other quantum computing physical implementation solutions.

Figure 3 Shallow neural network structure for quantum bit state reading and classification

This achievement was recently published in Physical Review Applied, a well-known international journal of applied physics. Dr. Duan Peng and Master Chen Zifeng from the Key Laboratory of Quantum Information of the Chinese Academy of Sciences are the co-first authors of the article, and Professor Guo Guoping is the corresponding author. This work was funded by the Ministry of Science and Technology, the National Natural Science Foundation of China, the Chinese Academy of Sciences and Anhui Province.