QuSquare: Scalable Quality-Oriented Benchmark Suite for Pre-Fault-Tolerant Quantum Devices

TL;DR

QuSquare introduces a scalable, hardware-agnostic benchmarking framework for pre-fault-tolerant quantum devices by coupling four workloads: Partial Clifford Randomized Benchmark, Multipartite Entanglement, TFIM Hamiltonian Simulation, and Data Re-uploading QNN. It defines rigorous design principles, concrete execution protocols, and reporting requirements to enable fair cross-platform comparisons, reproducibility, and progress toward standardized performance metrics. The suite combines scalable Clifford benchmarking with entanglement, quantum-simulation, and quantum-machine-learning tasks, supported by theoretical analyses of error cancellation, QSP circuit construction, and practical measurement strategies. Collectively, QuSquare provides a comprehensive, verifiable path toward evaluating and guiding the development of near-term quantum hardware and software ecosystems.

Abstract

As quantum technologies continue to advance, the proliferation of hardware architectures with diverse capabilities and limitations has underscored the importance of benchmarking as a tool to compare performance across platforms. Achieving fair, scalable and consistent evaluations is a key open problem in quantum computing, particularly in the pre-fault-tolerant era. To address this challenge, we introduce QuSquare, a quality-oriented benchmark suite designed to provide a scalable, fair, reproducible, and well-defined framework for assessing the performance of quantum devices across hardware architectures. QuSquare consists of four benchmark tests that evaluate quantum hardware performance at both the system and application levels: Partial Clifford Randomized, Multipartite Entanglement, Transverse Field Ising Model (TFIM) Hamiltonian Simulation, and Data Re-Uploading Quantum Neural Network (QNN). Together, these benchmarks offer an integral, hardware-agnostic, and impartial methodology to quantify the quality and capabilities of current quantum computers, supporting fair cross-platform comparisons and fostering the development of future performance standards.

Figures (15)

• Figure 1: Flowchart of the Partial Clifford Randomized Benchmark. The diagram summarizes the steps described in Protocol \ref{['protocol:2']}, including parameter initialization, circuit generation and execution, measurement estimation, and the iterative update of $\mu$ based on the estimated infidelity $r_\mu$.
• Figure 2: Flowchart of the Multipartite Entanglement Benchmark. The diagram summarizes the steps described in Protocol \ref{['GHZ_protocol']}, including the selection of Pauli strings, the construction and execution of the quantum circuit, the computation of the fidelity estimator, and the iterative increase of the number of qubits $N$.
• Figure 3: Quantum circuit representation of the QSP implementation used to approximate the time-evolution operator $e^{-i\hat{H}t}$. The block $\hat{U}_{\rm QSP}^{(\cos)}$ and $\hat{U}_{\rm QSP}^{(\sin)}$ denotes the quantum circuit used to implement $P_{\cos}^{\epsilon_0}((\alpha t)\hat{H}/\alpha)$ and $P_{\sin}^{\epsilon_0}((\alpha t)\hat{H}/\alpha)$ respectively using quantum signal processing. The evolved quantum state $|\psi(t)\rangle$ is obtained upon post-selection of the ancillary qubit $q_B$ in the state $|1\rangle$, while $q_A$ and all $m$ ancilla qubits in the state $|0\rangle$.
• Figure 4: Circuit representation of the QSP sequence described in Eq. \ref{['QSP']}, consisting of alternating phase-controlled projectors $\Pi_{\phi_k}$ with $k=0,\dots,d$ and the block-encoding unitary $\hat{U}_H$. The circuit uses a single ancilla qubit $q_A$ for controlled operations, $m$ ancillary qubits to implement the block encoding of the Hamiltonian, and the qubits encoding the system state $\left|{\psi_s}\right\rangle$.
• Figure 5: Flow chart of the TFIM Hamiltonian simulation benchmark following the steps described in Protocol \ref{['protocol:TFIM_protocol']}. The diagram illustrates the iterative procedure used to determine the maximum evolution time $t_{\max}$ for which the estimated total magnetization $\mu$ remains within the interval $C = [\, M_z(t) - 10(2\epsilon_0 + \varepsilon),\; M_z(t) + 10(2\epsilon_0 + \varepsilon)\,]$