Optimization of high-temperature superconducting multilayer films using artificial intelligence

Abstract

<p>We have studied the possibility of utilizing artificial intelligence (AI) models to optimize<br>high-temperature superconducting (HTS) multilayer structures for applications working in a<br>specific field and temperature range. For this, we propose a new vortex dynamics simulation<br>method that enables unprecedented efficiency in the sampling of training data required by the AI<br>models. The performance of several different types of AI models has been studied, including kernel<br>ridge regression (KRR), gradient-boosted decision tree (GBDT) and neural network. From these,<br>the GBDT based model was observed to be clearly the best fitted for the associated problem. We<br>have demonstrated the use of GBDT for finding optimal multilayer structure at 10 K temperature<br>under 1 T field. The GBDT model predicts that simple doped-undoped bilayer structures, where<br>the vast majority of the film is undoped superconductor, provide the best performance under the<br>given environment. The obtained results coincide well with our previous studies providing further<br>validation for the use of AI in the associated problem. We generally consider the AI models as<br>highly efficient tools for the broad-scale optimization of HTS multilayer structures and suggest<br>them to be used as the foremost method to further push the limits of HTS films for specific<br>applications.<br></p>