Optimisation of Pulse Waveforms for Qubit Gates using Deep Learning

TL;DR

This paper introduces a deep learning-based method to optimize pulse waveforms for qubit gates, significantly improving fidelity for single-qubit gates and providing insights into multi-qubit gate challenges.

Contribution

It presents a novel DNN approach for pulse optimization in quantum computing, enhancing gate fidelity and addressing multi-qubit gate complexities.

Findings

Achieved near-perfect fidelities for single-qubit gates (above 0.9999).

Demonstrated the method's potential for high-fidelity quantum gate implementation.

Identified challenges in multi-qubit gate fidelity due to entanglement effects.

Abstract

In this paper, we propose a novel method using Deep Neural Networks (DNNs) to optimise the parameters of pulse waveforms used for manipulating qubit states, resulting in high fidelity implementation of qubit gates. High fidelity quantum simulations are crucial for scaling up current quantum computers. The proposed approach uses DNNs to model the functional relationship between amplitudes of pulse waveforms used in scheduling and the corresponding fidelities. The DNNs are trained using a dataset of amplitude and corresponding fidelities obtained through quantum simulations in Qiskit. A two-stage approach is used with the trained DNNs to obtain amplitudes that yield the highest fidelity. The proposed method is evaluated by estimating the amplitude for pulse scheduling of single (Hadamard and Pauli-X) and two qubit gates (CNOT). The results clearly indicate that the method can achieve high fidelity implementations of single-qubit gates with fidelities of 0.999976 and 0.999923 for Hadamard and Pauli-X gates, respectively. For the CNOT gate, the best fidelity obtained is 0.695313. This can be attributed to the effects of entanglement and the need for the phase parameter to be accounted for within the predictive model.