Identifying optimal amorphous materials for fluoride removal through Monte Carlo and neural network modeling

Abstract

Capturing CF₄ is crucial for mitigating its substantial greenhouse effect and environmental impact in the microelectronics industry. Here we employed a hybrid approach combining grand canonical ensemble Monte Carlo molecular simulations and neural network models to screen over 100 amorphous materials for N₂/CF₄ gas adsorption storage and separation. Materials with higher adsorption capacities exhibited densities around 0.7 to 1.0 g/cm³ and pore sizes within the range of 1.4–1.6 Å. At 298 K and 1000 kPa, HCP-Colina-id0016 and aCarbon-Bhatia-id001 demonstrated the highest CF₄ adsorption, reaching 5.65 and 5.34 mmol/g, respectively. For the separation of N₂/CF₄ mixtures, considering the comprehensive CF₄ adsorption selectivity and capacity, we recommend HCP-Colina-id0016 at high pressure conditions (4500 kPa) and aCarbon-Bhatia-id001 at medium to low pressures (below 500 kPa). The separation of mixtures is more favorable at low CF₄ concentrations, becoming more challenging as CF₄ concentration increases. Additionally, the Ideal Adsorbed Solution Theory (IAST) accurately predicted the separation of the N₂/CF₄ system on amorphous materials. We found that the genetic algorithm-optimized neural network (GA-BP) outperformed the standalone backpropagation neural network (BP) in accurately predicting the relationship between material structural properties and CF₄ adsorption, showing its potential for widespread application in large-scale material screening.