Quantum computers represent a groundbreaking paradigm that promises to redefine the boundaries of infomation processing.[1] However, it is not clear whether a quantum computer can provide greater computational power at a lower energy cost. One of the issues is that a significant amount of power is consumed to calibrate qubits because each qubit has different control parameters that even change over time. To calibrate the qubits, quantum state tomography (QST) and quantum gate set tomography (QGST), aiming at reconstructing an unknown state via measurement has been recognized as a fundamental ingredient. To reconstruct a quantum state or a density of matrix, one should perform many measurements and gather statistical outcomes.

In near future, the number of quantum states and hence the measurements increase exponentially to achieve informational completeness for large-scale quantum systems. In this talk, I will discuss how AI and machine learning (ML) can effectively calibrate the qubits by reconstructing qubit states and hence reducing the number of measurement [2,3] and also propose that memristors operating at a cryogenic temperature [4] can be used for minimizing the power consumption of the qubit calibration, hence the total power consumption of the quantum computers. We introduced transfer models and the advanced training to enhance model performance, demonstrating QGST for 2 and 3 gates on single-qubit and two-qubit systems, with noise estimation with comparable accuracy to the conventional QGST.

[2] In order to scale these approaches to larger, practical quantum systems demands not only algorithmic efficiency but also energy-efficient hardware. In-memory-computing using resistive random-access memory (ReRAM), from which memristors are the most promising candidate, offers an energy efficiency in QST.

We have simulated spiking variation autoencoder (SVAE) using the spiking feature of memristors and have achieved a high reconstruction fidelity while requiring significantly fewer computational resources [3]. We also have developed interface-enhanced memristors tailored for cryogenic operation.[4] The cryo-memristor with the material stack of Pt/Ti/HfO2/Pt provides eight nonvolatile resistive levels with an ultra-low read noise rate of 0.3% at a temperature of 4K. The device is promising for realizing a cryogenic QGST hardware for qubit calibration for energy-efficient quantum computers.