Digital Twin

Qruise utilizes Machine Learning techniques to create a highly detailed white-box, physics based differentiable model of the quantum computer. This includes utilising Reinforcement Learning optimizers for gate calibration, active learning and Bayesian Optimal Experiment design, based on surrogate NN models to acquire the right data from which to determine the model parameters and additional novel techniques. Together, these techniques will allow to create a detailed predictive model of the quantum computer, including not just the quantum components, but also a detailed modelling of the control electronics and their imperfections – anything and everything which may affect overall device performance. Using this digital twin it will not only be possible to design noise-decoupling and noise-robust gates, but more importantly determine the specific factors in the device that currently limit its performance, allowing physicists to better focus their efforts on improving the device.

Optimal Control

The application of optimal control algorithms is essential to realize the actual possible performance of a given device. In quantum control theory, such algorithms are used to optimize the time-varying form of pulses or pulse sequences for classical electromagnetic fields that control the dynamics of a quantum system. Applied to quantum systems, this means the best possible execution of basic operations on the hardware level, such as quantum gates or the initialization or readout of information. Best possible refers to both the smallest possible error and the shortest possible duration of the corresponding operation. In fact, the optimized modulation of the pulses or pulse sequences often allows a much shorter operation duration than with manually developed pulses, up to the physical limit - the so-called quantum speed limit. Furthermore, optimal control algorithms can be used to determine pulses for hardware operations for which there is no simple mathematical solution. This is particularly relevant for complex multi-qubit cases and when the effect of control electronics/photonics is to be taken into account. Another important application of optimal control algorithms is the determination of pulse sequences that are able to compensate for spatial or temporal inhomogeneity due to a variance in the system parameters.

Teams

Teamwork has never been easier! We make it trivial to share within the lab and, should you choose, with other collaborators any jobs, experiments or data.

Hardware agnostic

Fully parallelisable and platform agnostic – works out of the box with Superconducting, Rydberg, NV-centres, Ion-trap QPUs and hybrid control stacks. Quickly and easily integrate with your current experiment's control electronics and software with effortless integration. Integration with experimental hardware is a core capability of the Qruise software stack. Whether you currently access the control electronics APIs directly, or through a software layer already present in the laboratory – either way, we can connect to it.

Flexible deployment

Qruise software supports a very flexible deployment architecture – fully in the cloud, full on-premise, and hybrid. You pick, we deploy

Lab private cloud

Qruise brings you the cloud in your lab. All the convenience features are there; treat multiple physical experiments and multiple digital twins as a cloudservice – sending “jobs” and receiving results

Closed loop calibration

As the digital twin is initially not very accurate, the control parameters must be further optimized using the quantum device itself, using a process called closed-loop calibration (a.k.a. tune-up). Don’t worry, you don’t need to turn any nobs! Qruise will do that for you automatically.

Bayesian experiment design

If the existing experimental data is not sufficient to determine the value of a model parameter, the software can be used togenerate the controls for a series of experiments, carry out the experiments, analyze the results, and then determine the parameter's value. This process utilizes Reinforcement Learning agents which predict policies to design optimal experiments for learning any arbitrary parameters about the quantum device.

Adversarial model validation

Qruise will optimize the design experiments so as to best measure system parameters, and proactively search for experimental scenarios which deviate from the optimized system model or desired behavior via adversial model validation. What are the benefits? Rapidaly identifying dataset shifts, enhance generalization, optimize resource allocation, improve model robustness, just to name a few.

Error budget (“What if”)

The root cause of performance bottlenecks: A detailed trace-back of the observed infidelity back to its originating physical phenomena (noise source, cross-talking elements, parasitic couplings concurrently affecting qubit dynamics). All these phenomena can be turned off in simulation one-at-a-time, to determine how much infidelity is caused by each. This is then actionable information, as it can guide the efforts in improving the hardware in the next iteration

Co-design optimization

Explore how device performance can be improved in the next generation of hardware by exploring new model regimes. One is not limited to optimizing controls, but may concurrently optimize both controls and device parameters (e.g. qubit couplings or AWG bandwidth), to determine the overall bestworking settings.

Automatic bring-up

Significantly reduce the time and effort required to bring a quantum device online.

Automatic recalibration

Unfortunately, quantum devices drift. Qruise provides the methods to perform ongoing automatic recalibration. This can be coupled to the model learning module, so that the system model is also updated continuously, giving you insight into exactly which physical parameters drift and how. This is not just increasing uptime, but also maintaining the performance and reliability of your system.

Custom dashboards

Dashboards and visualizations allow you to monitor the device’s status and parameters. Gate fidelities can also be communicated upstream to your compiler, to allow for fidelity-sensitive compilation and execution of quantum circuits. From data visualization, monitoring, control and interactions, alerts and notifications, data filtering and aggregation, access raw data, annotation and collaboration, and many more – in one place.

Model learning

Optimizing the model to best replicate observed experiment data. No coding of fitting functions is necessary with this approach, making it possible to determine multiple parameter values simultaneously. Moreover, Qruise can produce detailed 1D and 2D sensitivity plots to show how tightly bound are the model parameters given the experimental data, in terms of standard deviations.