Machine Learning-Assisted Efficient Discovery and Rational Design of Thermally Conductive Polymers

As electronic chip technology advances toward miniaturization, integration, and high-frequency capabilities, along with the continuous increase in device power density, the high thermal conductivity (TC) of polymers becomes increasingly important in thermal management for flexible electronics and optoelectronics. However, the poor thermal conduction of polymers is always an obstacle. To this end, this paper proposed a novel machine learning-assisted framework that simultaneously realizes the rapid screening and rational design of polymers with high thermal conductivity (TC > 0.40 W m–1 K–1). The deep neural network model was trained to map the relationships between the microstructures and TCs of polymers, and it enabled high-throughput screening for highly thermally conductive candidates in the predefined exploration space. Further, we extracted some promising fragments from the discovered polymers and designed a series of innovative, highly thermally conductive structures by combining the Monte Carlo tree search algorithm and molecular generation rules. Ultimately, we demonstrated the beneficial impact of chain stiffness on TC, relating it to the chain conformation and the distribution of bond strengths in the amorphous systems. This two-pronged strategy, transitioning from discovery toward design, provides a rational and efficient avenue for accelerating the exploitation of polymers with high TC and other desired properties.