SynthVANet for Effective Prediction of Vertically Aligned Nanocomposite Synthesis

The creation of self-assembled vertically aligned nanocomposite (VAN) thin films via a one-step pulsed laser deposition process has garnered tremendous research interests due to their multifunctionality and unique features such as vertical strain control and anisotropic structure. However, synthesizing VAN structures poses significant challenges, often requiring a trial-and-error approach because of their complex mechanics. To address this, we have developed an AI-driven tool to help guide the synthesis of VAN efficiently. This method automatically extracts data from published literature via a large language model (LLM), ChatGPT, combined with lab-collected data to create a comprehensive database. The crystal-graph-based neural network, SynthVANet, utilizes this database for binary classification of VAN and non-VAN structures, achieving high accuracy as validated by precision, recall, and F1 scores. Overall, this model aims to offer researchers a means to assess the probability of success associated with their growth recipes, significantly reducing experimental time and costs.