Machine Learning-Guided Design of Bio-Based Furan-Containing Polyimides with Tunable Dielectric Properties

The development of biobased furan-containing polyimides with tailored dielectric properties is critical for sustainable electronics, yet their structure–property relationships remain underexplored. Here, we combine density functional theory (DFT) calculations and machine learning to predict the dielectric performance of furan-based polyimides and systematically compare them with conventional analogues. We first employed high-throughput computational techniques based on DFT and density functional perturbation theory to establish a high-frequency dielectric constant data set comprising 3032 furan-based polyimides and 2061 phenyl-based polyimides. Using this data set, we trained a graph neural network model, which achieved an R 2 greater than 0.82 in predicting the average high-frequency dielectric constant on the test set. A comparison of the model’s predictions with existing experimental results showed similar trends, further validating the predictive accuracy of our model. Then we evaluated the impact of furan substitution for phenyl groups on the dielectric constant across the entire chemical space, achieving wider dielectric constant distribution of furan-containing polyimides. Our study reveals the importance of furan-based polymers in the application of electronics.