Neural network search for optimal pulse trains for qubit dynamics control
Michael A. Sergeev, M. V. Bastrakova, Vsevolod A. Vozhakov, N. V. Klenov, Igor I. Soloviev, Maxim V Tereshonok · T-Comm - Телекоммуникации и Транспорт · 2024
Despite its insensitivity to charge noise, the transmon's anharmonicity and control pulse length are limited. Transmon state control traditionally involves a quadrature mixer that mixes microwave signals from a room-temperature oscillator and an arbitrary waveform generator to control the single qubit states. Scaling upquan tum processors faces challenges in hardware, management of qubit operations, and read-out procedure due to the large amount of expensive room-temperature equipment required for each qubit. Operating at millikelvin temperatures, these devices introduce thermal noise, reducing qubit lifetime and distorting control signals. An alternative promising control method is based on superconducting digital electronics. In these digital circuits, a bit of information is represented by a short unipolar voltage pulse generated when a single flux quantum (SFQ) pulse passes through a Josephson junction. The qubit states are controlled by the action of a sequence of SFQ pulses, with the pulse-to-pulse timing adjusted to induce a coherent rotation of the state vector in the computational subspace and to minimize leakage to the outside. The paper discusses a method for controlling the states of a transmon qubit using digital superconducting electronics. In this approach, the sequences of picosecond voltage pulses are used to control the state of a quantum computing system. We have considered a control scheme based on a bipolar pulse generator and proposed an algorithm for finding the optimal implementation of a bipolar short pulse control sequence for performing high-precision single-bit operations (with fidelity equal to 99,99 %) using deep learning algorithms with rein forcement: AlphaGo Zero, AlphaZero and Proximal Policy.