The SpinPulse library for transpilation and noise-accurate simulation of spin qubit quantum computers

TL;DR

SpinPulse presents a modular, open-source framework for pulse-level simulation of spin-qubit quantum computers that explicitly models non-Markovian noise to enable realistic gate fidelity assessments and error mitigation. By translating circuits into a native gate set, converting them to time-dependent pulse sequences, and integrating them with a noise environment, SpinPulse provides end-to-end simulations that reflect hardware-specific dynamics. The package supports dynamical decoupling, gate-exchange noise, and tensor-network simulations via quimb, enabling scalable studies of large circuits and clusters. This work offers a practical tool for hardware-aware quantum circuit design and analysis, with extensible components for future improvements in pulse shaping, connectivity, and ESR-based control.

Abstract

We introduce SpinPulse, an open-source python package for simulating spin qubit-based quantum computers at the pulse-level. SpinPulse models the specific physics of spin qubits, particularly through the inclusion of classical non-Markovian noise. This enables realistic simulations of native gates and quantum circuits, in order to support hardware development. In SpinPulse, a quantum circuit is first transpiled into the native gate set of our model and then converted to a pulse sequence. This pulse sequence is subsequently integrated numerically in the presence of a simulated noisy experimental environment. We showcase workflows including transpilation, pulse-level compilation, hardware benchmarking, quantum error mitigation, and large-scale simulations via integration with the tensor-network library quimb. We expect SpinPulse to be a valuable open-source tool for the quantum computing community, fostering efforts to devise high-fidelity quantum circuits and improved strategies for quantum error mitigation and correction.

Figures (5)

• Figure 1: Organization of our package with the three steps (1), (2), (3) described in Sec. \ref{['sec:organization']}. A qiskit circuit is converted to an Instruction Set Architecture (ISA) circuit using an instance of HardwareSpecs class, which is equipped with a hardware model specific qiskit transpiler. This ISA circuit is then converted to a PulseCircuit instance that describes a pulse-level circuit in the presence of additional noise terms. This circuit can then be numerically integrated to form a noise-accurate qiskit circuit.
• Figure 2: Examples of time traces $\epsilon(t)$ for a) quasi-static, b) white, and c) pink noise, for the same value of $T_2^*=100$, duration $2^{18}$. For quasi-static noise, we use time traces of $\texttt{segment\_duration}=250$. In panels d),e), f), we show the corresponding averaged Ramsey contrast calculated using the method plot_ramsey_contrast.
• Figure 3: Dynamical decoupling for the first three layers of the circuit shown in Fig. \ref{['fig:package']}. With a) spin echo, only the second idle pulse sequence can accommodate the two $X$ pulses. With b) full drive, this is replaced by an $R_{X}(2n\pi)$ rotation whose amplitude is adjusted to occupy the full duration of the sequence. Here, $n=1$.
• Figure 4: Gate channels in the $\chi$ matrix form for idle (top), $X$ (middle) and $R_{ZZ}$ (bottom) gates as shown using the method plot_chi_matrix. We use the three types of noise, with the (many) parameters shown in the file AverageSuperoperatorsNoisyGates.py. The analytical expressions are written in the model section \ref{['sec:model']}.
• Figure 5: State fidelity for a cluster state prepared within our model, in the presence of pink noise of coherence time $T_2^*$ and for various qubit numbers $N$. All the other simulation parameters are presented in the dedicated Jupyter notebook.