AI-Driven Discovery and Molecular Engineering Design for Enhancing Interface Stability of Black Phosphorus

Molecular engineering offers significant potential for developing advanced interfacial materials, yet the complexity of organic molecules poses challenges in discovering optimal structures. This study leveraged large language model (LLM) and machine learning (ML) to accelerate molecular discovery and guide molecular engineering for enhancing the stability of black phosphorus (BP), a promising 2D semiconductor but rapidly degrades when exposed to oxygen and moisture. By utilizing GPT-4o, molecular groups such as ─SiR₃, ─PR₂, ─SH, and ═NH that interact effectively with BP were identified and a high-throughput workflow employing graph neural networks (GNNs) models was developed to successfully predict and screen 662 promising candidates from over 117 million molecules. These candidates were validated by density functional theory (DFT) simulations and experiments, with synthesis protocols guided by GPT-4o, achieving great interfacial stabilization of BP for up to 24 days under ambient conditions. Furthermore, a new synergistic molecular engineering strategy was proposed by incorporating functional head, linker, and tail groups of molecules to even enable the use of hydrophilic molecules to stabilize BP surface, overcoming traditional design limitations. This work highlights the AI technologies not only in optimizing BP interfacial stability but also in broader aspects of molecular engineering for various materials.