The University of Oxford has spearheaded a groundbreaking study that harnesses the capabilities of machine learning to tackle a significant challenge in the realm of quantum devices. This pioneering research, detailed in Physical Review X, marks the first successful attempt to bridge the “reality gap” – the variance between predicted and observed behaviors in quantum devices.
Quantum computing holds immense potential for applications ranging from climate modeling and financial forecasting to drug discovery and artificial intelligence. However, the scalability and integration of individual quantum devices, or qubits, pose a considerable hurdle due to inherent variability. Even seemingly identical units exhibit diverse behaviors, attributed to nanoscale imperfections in the materials constituting quantum devices.
The study employed a “physics-informed” machine learning approach to indirectly deduce these disorder characteristics based on their impact on electron flow through the device. Associate Professor Natalia Ares, the lead researcher from the Department of Engineering Science at the University of Oxford, likened the process to improving predictions in “crazy golf” using simulations, machine learning, and additional shots.
Researchers measured the output current across an individual quantum dot device at various voltage settings, inputting the data into a simulation. By calculating the disparity between the measured and theoretical current in the absence of internal disorder, the simulation identified disorder arrangements explaining the measurements across different voltage settings. This innovative approach seamlessly integrated mathematical, statistical methods, and deep learning.
Ares elaborated on the analogy, comparing it to placing sensors along a tunnel in crazy golf to measure the ball’s speed at different points, improving predictions despite being unable to directly observe internal disorder.
The model not only identified suitable internal disorder profiles but also accurately predicted voltage settings for specific device operating regimes. This breakthrough offers a novel means of quantifying variability between quantum devices, enhancing predictions of device performance and aiding in the engineering of optimal materials for quantum applications. The study also introduces potential compensation approaches to mitigate the effects of material imperfections in quantum devices.
David Craig, a Ph.D. student at the Department of Materials, University of Oxford, drew a parallel between the study and inferring the presence of black holes from their effects on surrounding matter. He emphasized the utility of using physics-aware machine learning to narrow the reality gap, acknowledging the model’s ability to address the inherent complexity of real quantum devices.
Source: University of Oxford