Quantum machine learning based on multimode superconducting piezomechanics

Developing a hardware-efficient quantum platform consisting of multimode piezomechanics strongly coupled with superconducting qubits for quantum machine learning applications.

A schematic of the multimode piezomechanical QML platform. A bulk acoustic resonator provides a large number of bosonic modes (green balls) for multidimensional data encoding. A superconducting qubit (red) is strongly coupled with the acoustic modes to provide nonlinearity and quantum state control.

Quantum information science (QIS) and artificial intelligence (AI) are two revolutionary information technologies. Machine learning (ML), a key branch of AI, is now indispensable for a variety of applications. On the other hand, quantum computing exploits quantum phenomena to achieve substantial computational speed-ups over classical devices. Combining the most advanced QIS and AI technologies, quantum ML (QML) is an emerging, exciting field for addressing previously untenable computational tasks.

Encoding data in quantum states, QML harnesses quantum correlations in a large Hilbert space to efficiently perform computations. To fully exploit the advantages of QML, a hardware platform that supports a large number of quantum states is highly desired. Despite recent progress toward large-scale integrated superconducting qubits, the hardware cost and overhead for preparing many quantum states using such two-level systems are intrinsically high. It is still a tremendous challenge to realize a hardware-efficient quantum architecture for QML testbeds.

At the Center for Nanoscale Materials at Argonne, scientists are developing a novel hardware-efficient multimode piezomechanical system for QML. The system features many highly coherent bosonic modes in a piezo bulk acoustic resonator [1, 2] for multidimensional data encoding. By strongly coupling these bulk acoustic modes with superconducting qubits, strong nonlinearity can be obtained to create quantum correlations among the encoded bosonic modes for QML tasks [3].

In contrast to qubit systems, a bosonic mode provides an infinitely large Hilbert space, key for advanced hardware-efficient quantum encoding techniques. Moreover, the multimode mechanical resonator has an ultracompact footprint, orders of magnitude smaller than superconducting resonators, further enhancing hardware efficiency. This hardware-efficient multimode quantum system could serve as an essential platform for developing quantum advantages in ML tasks such as classification and regression.

References:

[1] X. Han, C.-L. Zou, and H. X. Tang, Multimode strong coupling in superconducting cavity piezoelectromechanics, Physical Review Letters 117, 123603 (2016).

[2] X. Han, C.-L. Zou, W. Fu, M. Xu, Y. Xu, and H. X. Tang, Superconducting Cavity Electromechanics: The Realization of an Acoustic Frequency Comb at Microwave Frequencies, Physical Review Letters 129, 107701 (2022).

[3] J. Liu, C. Zhong, M. Otten, A. Chandra, C. L. Cortes, C. Ti, S. K. Gray, and Xu Han, Quantum Kerr Learning, Machine Learning: Science and Technology 4, 025003 (2023).