Deep Learning Augmented Topologically-Protected Photocatalysts

ended

Sunlight, as a non-stop power source granted by nature, provides about ten thousand times more energy than humans consume globally. Therefore, its harvesting and conversion to storable energy, such as plants perform in photosynthesis, represents a long-held dream of humanity. With the rapid progress of photocatalysis, humankind now endeavors to split water molecules using sunlight, thus storing solar energy into clean and recyclable hydrogen gas. To date, the efficiency of this conversion is up to 20% but with insufficient stability. In this context, DELATOP represents an effective solution to boost solar-to-hydrogen (StH) efficiency while significantly improving conversion robustness. Recently, cavity chemistry has arisen as a novel path to control chemical reaction rates in the context of light-matter interactions. Concurrently, photonic devices with topologically protected resonances have demonstrated superior defect tolerance and life-cycle durability. In this regard, DELATOP aims to design novel photocatalytic heterojunctions endowed with exceptional photon harvesting and carrier generation rate. Furthermore, using artificial intelligence (AI) for reverse engineering design, the R&D cycles can be significantly reduced with proper optimizations. As a result, the first AI-designed topo-photocatalysts will be delivered, conjugating high-imperfection tolerance and a super-extended lifetime of photo-carriers (~100 times), i.e., smart management of photons and carriers for the next-generation of green energy technologies. The project identifies three objectives to reach the final goal: I) Conceive and design novel photonic solutions based on topologically-protected resonances to be applied in the photocatalytic context; II) Deliver the first AI-designed topo-photocatalyst through injecting deep learning neurons into the previous design; III) Fabrication and characterization of topologically protected photocatalytic devices with enhanced StH conversion efficiency.